{"uuid": "73027e11-3572-4315-8f99-466230aad030", "vulnerability_lookup_origin": "1a89b78e-f703-45f3-bb86-59eb712668bd", "author": "9f56dd64-161d-43a6-b9c3-555944290a09", "vulnerability": "CVE-2026-105753", "type": "seen", "source": "https://gist.github.com/alon710/c6bdbc8fe870bf2cbcc1807d8a196cec", "content": "# CVE-2026-105753: CVE-2026-105753: Reachable Assertion in vLLM Multimodal IPC Cache Leading to Denial of Service\n\n> **CVSS Score:** 6.5\n> **Published:** 2026-10-06\n> **Full Report:** https://cvereports.com/reports/CVE-2026-105753\n\n## Summary\nA state desynchronization (cache drift) vulnerability exists in the multimodal Inter-Process Communication (IPC) Least Recently Used (LRU) caches of vLLM. When a multimodal request fails validation after its media hash has been registered on the frontend but before the payload is committed to the backend engine core, the frontend and backend caches drift out of lockstep. A subsequent request reusing the same media triggers an assertion failure in the backend engine core, resulting in a complete denial of service.\n\n## TL;DR\nA state desynchronization between vLLM's frontend and backend caches allows remote authenticated users to trigger a fatal assertion crash (CWE-617) by submitting a validation-failing multimodal request followed by a duplicate request, leading to complete server denial of service.\n\n## Exploit Status: POC\n\n## Technical Details\n\n- **CWE ID**: CWE-617 (Reachable Assertion)\n- **Attack Vector**: Network (AV:N)\n- **CVSS Score**: 6.5\n- **Impact**: Complete Denial of Service (DoS)\n- **Exploit Status**: Proof of Concept (PoC) / Unit Tests available\n- **KEV Status**: Not listed in CISA KEV\n\n## Affected Systems\n\n- vLLM Multimodal Serving Engine\n- **vllm**: < 0.28.0 (Fixed in: `0.28.0`)\n\n## Mitigation\n\n- Upgrade vLLM to version 0.28.0 or later to apply the official security patches.\n- Disable multimodal IPC caching if the serving workload does not require repetitive media processing.\n- Enforce prompt and token limit validation at an upstream proxy or API gateway before requests reach the vLLM instance.\n\n**Remediation Steps:**\n1. Identify all deployed containers or environments running vLLM versions prior to 0.28.0.\n2. Update the deployment dependencies or container base images to reference vLLM >= 0.28.0.\n3. Restart the vLLM inference service and verify that the warning logs for cache misses resolve gracefully on retryable inputs.\n\n## References\n\n- [GitHub Security Advisory GHSA-ph3r-5jfg-f84f](https://github.com/vllm-project/vllm/security/advisories/GHSA-ph3r-5jfg-f84f)\n- [NVD CVE-2026-105753 Detail](https://nvd.nist.gov/vuln/detail/CVE-2026-105753)\n- [vLLM Pull Request #46747](https://github.com/vllm-project/vllm/pull/46747)\n- [vLLM Pull Request #51897](https://github.com/vllm-project/vllm/pull/51897)\n\n\n---\n*Generated by [CVEReports](https://cvereports.com/reports/CVE-2026-105753) - Automated Vulnerability Intelligence*", "creation_timestamp": "2026-10-06T02:30:31.000000Z"}