CVE-2026-22773: vLLM is vulnerable to DoS in Idefics3 vision models via image payload with ambiguous dimensions
(updated )
Users can crash the vLLM engine serving multimodal models that use the Idefics3 vision model implementation by sending a specially crafted 1x1 pixel image. This causes a tensor dimension mismatch that results in an unhandled runtime error, leading to complete server termination.
References
- github.com/advisories/GHSA-grg2-63fw-f2qr
- github.com/pypa/advisory-database/tree/main/vulns/vllm/PYSEC-2026-143.yaml
- github.com/vllm-project/vllm/commit/0ec84221718d920c3f46da879cc354f94b8fb59e
- github.com/vllm-project/vllm/pull/29881
- github.com/vllm-project/vllm/security/advisories/GHSA-grg2-63fw-f2qr
- nvd.nist.gov/vuln/detail/CVE-2026-22773
Code Behaviors & Features
Detect and mitigate CVE-2026-22773 with GitLab Dependency Scanning
Secure your software supply chain by verifying that all open source dependencies used in your projects contain no disclosed vulnerabilities. Learn more about Dependency Scanning →