MediumVulnerabilityLLM-specific
CVE-2026-105753: vLLM denial of service via multimodal cache hash reuse
- Source
- NVD(opens in a new tab)
- Identifier
- CVE-2026-105753
- Published
- Record updated
- Affected
- vllm < 0.28.0
- Fixed in
- 0.28.0
- released August 26, 2026, 40 days before disclosure
- Known exploitation
- Not listed in the CISA Known Exploited Vulnerabilities catalog at the last check.
- EPSS
- 0.4%
Summary
CVE-2026-105753 affects vLLM, an inference and serving engine for large language models, prior to 0.28.0. The default mirrored multimodal LRU cache can record a media hash in the frontend sender cache before engine admission, while the engine receiver cache never gets the payload if the request is rejected. A later request reusing that media hash makes the receiver cache hit an assertion, causing an availability failure in the shared service.
Mitigation
Fixed in version 0.28.0.
Affected packages in the Exposure Registry
Matched by package name and ecosystem. Each entry shows whether the package delegates to a language model and how many tracked packages depend on it.
- vllmPyPILLM dependency since 2023-08-25 · 15 tracked dependents
Topics
Related items
- LowToken fraud is here and it’s BAD (YouTube)Similar attack · Simon Willison's Weblog
- HighCVE-2026-133161: Pydantic AI Harness ToolOutputLimits ANSI stripping slowed by quadratic regexSimilar attack · NVD/CVE Database
- HighCVE-2026-130686: vLLM resource exhaustion through many remote image URLs in one requestSimilar attack · NVD/CVE Database
- HighCVE-2026-130685: vLLM denial of service through out-of-range prompt token ids in /v1/completionsSimilar attack · NVD/CVE Database
- LowRented Oscillations: Multi-Tenant Data Center Workload Synchronization Attacks and MitigationSimilar attack · Arxiv (cs.CR + cs.CL + cs.LG)