Vllm-project vulnerabilities
125 CVEs and advisories name Vllm-project as an affected vendor, newest first. Follow it as a feed.
- CVE-2026-105758vLLM: Qwen2-VL / Qwen3-VL video samplers bound on request-controlled max_frames, which the num_frames ceiling does not reach
- CVE-2026-105757vLLM: Structured-output request errors escape the request boundary and terminate the shared EngineCore — engine-fatal denial of service (3 sites)
- CVE-2026-105756vLLM: Loose `cache_salt` validation lets a single request kill EngineCore on LMCache-MP deployments — uncaught downstream `ValueError` denial of service
- CVE-2026-105755vLLM: Flash late-interaction scoring caches query embeddings under a caller-controlled request id — cross-request integrity break and induced errors on `/score` and `/rerank`
- CVE-2026-105754vLLM: Scale-out disaggregated multimodal transport trusts caller-supplied features
- CVE-2026-105753vLLM: Mirrored multimodal IPC caches desync after a rejected request — a later request reusing the same media hash trips a receiver assertion in the engine core
- CVE-2026-105752vLLM: Harmony tool continuations drop `cache_salt` — restoring a cross-tenant prefix-cache membership oracle
- CVE-2026-103241vllm-project vLLM Gemma4UnifiedParser gemma4.rs denial of service
- CVE-2026-100654vLLM before 0.29.0 Denial of Service via out-of-range stop_token_ids
- CVE-2026-100653vLLM 0.22.1 before 0.28.0 Incomplete Artifact Pin Propagation
- CVE-2026-100652vLLM 0.22.0 through 0.23.0 Denial of Service via stop_token_ids
- CVE-2026-100651vllm before 0.29.0 Denial of Service via Decoder Prompt Length Bypass
- CVE-2026-100650vLLM before 0.29.0 Resource Exhaustion via Unbounded Media Materialization
- CVE-2026-100649vLLM before 0.29.0 Resource Limit Bypass via Sampler Subclass
- CVE-2026-100648vllm before 0.29.0 Uncontrolled Resource Consumption via Audio Decoding
- CVE-2026-100647vLLM before 0.29.0 CPU Exhaustion via unbounded cache_salt
- CVE-2026-94627vLLM through 0.29.0 GPU KV Cache Leak via Mooncake Transfer ID Collision
- CVE-2026-94626vLLM through 0.29.0 Memory Exhaustion via Unvalidated NIXL tp_size
- CVE-2026-94625vLLM through 0.29.0 Resource Exhaustion via Ownerless Mooncake Transfer Placeholders
- CVE-2026-94624vLLM through 0.29.0 Denial of Service via Unbounded P2P KV Offloading Sessions
- CVE-2026-94623vLLM through 0.29.0 Denial of Service via NIXL Multi-Prompt Assertion Failure
- CVE-2026-94622vLLM through 0.29.0 Denial of Service via Incomplete NIXL KV Transfer Metadata
- CVE-2026-93989vLLM through 0.29.0 Cross-Request Logits Corruption via bad_words
- CVE-2026-93841vLLM through 0.29.0 Adjacent-Request Sampler State Corruption via Unvalidated Prompt Token IDs
- CVE-2026-93840vLLM before 0.29.0 Cross-Request Logits Corruption via allowed_token_ids
- CVE-2026-93592vLLM before 0.28.0 Denial of Service via negative token ID
- CVE-2026-93436vLLM through 0.29.0 Memory Exhaustion via Rejected Requests
- CVE-2026-69147vLLM: Request-selected PyNvVideoCodec GPU decode bypasses static VRAM reservation
- CVE-2026-57173vLLM: Unauthenticated audio decompression-bomb DoS in /v1/chat/completions
- CVE-2026-92365vllm-project vllm thinking_budget_state.py algorithmic complexity