| /v1/systemone bypasses the --trust-request-chat-template gateGHSA-xgh8-qwjh-q853 | Medium | >= 0.32.0rc0, < 0.34.0 | 0.34.0 | 2026-11-18 |
| Thinking-budget sparse state aliases onto a co-batched request during a mixed `SWAP`, forcing the reasoning-end token into another request'sGHSA-fpch-3xrf-7q96 | Medium | >= 0.21.0, < 0.28.0 | 0.28.0 | 2026-11-09 |
| Cross-user token data leakage in async scheduling + Pipeline ParallelismGHSA-v97p-wpxp-px84 | Medium | >= 0.16.0, < 0.21.0 | 0.21.0 | 2026-11-09 |
| Response amplification in OpenAI validation error handlerGHSA-mwgq-9582-w2vf | Medium | >= 0.3.0, < 0.30.0 | 0.30.0 | 2026-11-07 |
| Missing request-wide media budget allows single-request frontend resource exhaustionGHSA-5r3c-5x4v-8gjh | High | >= 0.6.0, < 0.33.0 | 0.33.0 | 2026-11-02 |
| Unauthenticated EngineCore crash via out-of-range prompt token id under skip_tokenizer_init (incomplete fix of GHSA-25q3-v2hm-8vpf)GHSA-q9v9-qpjp-fc24 | High | >= 0.28.0, < 0.33.0 | 0.33.0 | 2026-11-02 |
| vLLM Speculators Config verifier.name_or_path Element Directory Traversal Information Disclosure VulnerabilityGHSA-qrj2-j723-m6w8 | Medium | < 0.33.0 | 0.33.0 | 2026-11-02 |
| vLLM: DoS via unbounded chat_template Jinja2 rendering exhausts the renderer thread poolCVE-2026-115911 | Medium | >= 0.19.0, < 0.32.0 | 0.32.0 | 2026-10-28 |
| vLLM: Public multimodal SHM handle forgery reaches worker-side `pickle.loads()` with insecure serialization disabledCVE-2026-115912 | High | < 0.32.0 | 0.32.0 | 2026-10-28 |
| vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engineCVE-2026-115913 | Medium | < 0.32.0 | 0.32.0 | 2026-10-28 |
| vLLM DeepSeek-OCR image processing vulnerable to denial of serviceCVE-2026-115914 | Medium | >= 0.11.1, < 0.32.0 | 0.32.0 | 2026-10-28 |
| vLLM: Untrusted EPS inputs can trigger denial of service or Ghostscript RCECVE-2026-115915 | High | < 0.32.0 | 0.32.0 | 2026-10-28 |
| vLLM: Out-of-vocabulary prompt token id passes input validation and triggers a CUDA device-side assert (engine crash / DoS)CVE-2026-115916 | Medium | < 0.32.0 | 0.32.0 | 2026-10-28 |
| vLLM: Permissive default CORS origins allow cross-origin requests to the vLLM HTTP serverCVE-2026-115917 | Low | < 0.32.0 | 0.32.0 | 2026-10-28 |
| vLLM: Responses API bypasses trust_request_chat_template for chat_template_kwargsCVE-2026-115918 | Medium | >= 0.16.0, < 0.32.0 | 0.32.0 | 2026-10-28 |
| vLLM denial of service via prompt embeds on XD-RoPE models (incomplete fix of CVE-2026-55514)GHSA-5q78-j82c-2vr2 | Medium | >= 0.28.0, < 0.30.0 | 0.30.0 | 2026-10-27 |
| vLLM OpenCVDynamicOpenPanguVideoBackend: request-level num_frames without a ceiling — unauthenticated memory/CPU exhaustion (OOM-kill) on anGHSA-894j-wg8m-63h8 | Medium | >= 0.30.0, < 0.31.0 | 0.31.0 | 2026-10-27 |
| Quadratic Cohere2 Vision preprocessing DoS via request-controlled max_patchesGHSA-9g3c-x747-c6wr | Medium | >= 0.10.1, < 0.31.0 | 0.31.0 | 2026-10-27 |
| vllm p2p never connected zmq_exhaustionCVE-2026-94624 | Medium | < 0.32.0 | 0.32.0 | 2026-10-27 |
| vLLM: Request-controlled mm_processor_kwargs.code_revision allows remote code executionCVE-2026-114053 | High | >= 0.7.3, < 0.31.0 | 0.31.0 | 2026-10-27 |
| sanitize_message() address regex is lowercase-only and does not match PIL's uppercase __repr__, leaking live heap addresses to unauthenticatGHSA-m9fg-w9qh-67pc | Medium | >= 0.14.0rc1, < 0.31.0 | 0.31.0 | 2026-10-27 |
| Incomplete CVE-2026-34755 remediation can be bypassed by request-level num_framesGHSA-vxqj-p4gw-9h4c | Medium | >= 0.19.0, < 0.31.0 | 0.31.0 | 2026-10-27 |
| MiniMax M3 video preprocessing permits request-driven CPU resource exhaustion through min_pixelsGHSA-w4cg-mhc7-mgw3 | Medium | >= 0.24.0, < 0.31.0 | 0.31.0 | 2026-10-27 |
| vLLM : DoS via unvalidated mm_processor_kwargs in multimodal image processors (Isaac, InternVL, H2OVL, etc.)CVE-2026-114055 | High | >= 0.7.3, < 0.31.0 | 0.31.0 | 2026-10-23 |
| vLLM: Forged multimodal UUID + P0/P1 processor-cache eviction drift: unauthenticated remote crash of the whole vLLM engineCVE-2026-114050 | Medium | < 0.31.0 | 0.31.0 | 2026-10-23 |
| Remote DoS in disaggregated multimodal /inference/v1/generate via malicious kwargs_data / mm_placeholdersGHSA-34qw-q2h2-jp3m | Medium | >= 0.20.0, < 0.30.0 | 0.30.0 | 2026-10-09 |
| Frame-count cap bypass in VideoMediaIO.load_base64 for video/jpeg via media_io_kwargs.video.num_framesGHSA-4m6g-h9mv-wqjc | High | >= 0.18.0, < 0.31.0 | 0.31.0 | 2026-10-09 |
| Rust gRPC output_candidates.token_ids bypasses logprob_token_ids limits and can terminate EngineCoreGHSA-6pgw-f3pv-4h8q | Medium | >= 0.22.0, < 0.24.0 | 0.24.0 | 2026-10-09 |
| Remote DoS in vLLM via allowed_token_ids Speculative-Decoding Metadata MismatchGHSA-8c65-hq7q-r7jm | High | >= 0.11.1, < 0.18.0 | 0.18.0 | 2026-10-09 |
| Nemotron-VL processors disable Pillow decompression-bomb guard globally, enabling unauthenticated DoSGHSA-mm38-5g96-7j6x | High | <= 0.8.5 | 0.25.0 | 2026-10-09 |
| Multimodal chat audio decoding bypasses the decompression-bomb duration limitGHSA-r45p-7qwx-qwjw | Medium | >= 0.19.0, < 0.24.0 | 0.24.0 | 2026-10-09 |
| Cross-tenant DoS via uninitialized LogprobsTensors with prompt_logprobsGHSA-rhcx-5729-88vg | High | >= v0.8.5, < 0.31.0 | 0.31.0 | 2026-10-09 |
| Rust gRPC CandidateTokens.top_n bypasses max_logprobs and can terminate EngineCoreGHSA-rwfw-xvg3-3937 | Medium | >= 0.22.0, < 0.24.0 | 0.24.0 | 2026-10-09 |
| SSRF via X-Request-Id header drives P2pNcclConnector outbound ZMQ+NCCL pair, leaking KV-cache tensorsGHSA-vfp2-c8pq-v6h6 | High | >= 0.9.2, < 0.24.0 | 0.24.0 | 2026-10-09 |
| Rust gRPC Generate accepts out-of-vocabulary prompt token IDs and can terminate EngineCoreGHSA-w2v2-5662-8v5q | Medium | >= 0.22.0, < 0.24.0 | 0.24.0 | 2026-10-09 |
| Unvalidated stop_token_ids causes engine crash (DoS) and out-of-bounds GPU writeGHSA-w9r6-29q4-pvpr | Medium | >= 0.20.1, < 0.29.0 | 0.29.0 | 2026-10-09 |
| An oversized min_tokens in one request hangs the vLLM engine while /health stays greenGHSA-4xqp-c3mv-qff7 | Medium | < 0.31.0 | 0.31.0 | 2026-10-06 |
| V1 InputBatch.condense leaves a stale allowed_token_ids mask on recycled batch rowsGHSA-6cxc-2vcg-w5qc | Low | < 0.31.0 | 0.31.0 | 2026-10-06 |
| MOSS-Audio request-controlled processor cache permits remote memory exhaustionCVE-2026-114052 | High | > 0.24.0, < 0.31.0 | 0.31.0 | 2026-10-06 |
| vLLM: Harmony tool continuations drop `cache_salt` — restoring a cross-tenant prefix-cache membership oracleCVE-2026-105752 | Low | < 0.30.0 | 0.30.0 | 2026-10-06 |
| vLLM: Mirrored multimodal IPC caches desync after a rejected request — a later request reusing the same media hash trips a receiver assertioCVE-2026-105753 | Medium | < 0.28.0 | 0.28.0 | 2026-10-06 |
| Caller-controllable values become or collide with shared cache keys, enabling cross-request cache aliasing (2 sites: multimodal EXIF hash anGHSA-rh6f-3x46-j33q | Medium | < 0.31.0 | 0.31.0 | 2026-10-06 |
| vLLM: Loose `cache_salt` validation lets a single request kill EngineCore on LMCache-MP deployments — uncaught downstream `ValueError` deniaCVE-2026-105756 | Medium | < 0.30.0 | 0.30.0 | 2026-10-05 |
| vLLM: Flash late-interaction scoring caches query embeddings under a caller-controlled request id — cross-request integrity break and induceCVE-2026-105755 | Medium | < 0.30.0 | 0.30.0 | 2026-10-05 |
| vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustionCVE-2026-105760 | Medium | >= 0.23.0rc2, < 0.30.0 | 0.30.0 | 2026-10-05 |
| vLLM: Structured-output request errors escape the request boundary and terminate the shared EngineCore — engine-fatal denial of service (3 sCVE-2026-105757 | Medium | < 0.30.0 | 0.30.0 | 2026-10-05 |
| vLLM: Scale-out disaggregated multimodal transport trusts caller-supplied featuresCVE-2026-105754 | Medium | < 0.30.0 | 0.30.0 | 2026-10-05 |
| vLLM: Qwen2-VL / Qwen3-VL video samplers bound on request-controlled max_frames, which the num_frames ceiling does not reachCVE-2026-105758 | Medium | >= 0.24.0, < 0.30.0 | 0.30.0 | 2026-10-05 |
| Crafted IAMF audio upload reaches a PyAV/FFmpeg native heap overflow through the speech transcription path — denial of serviceGHSA-m52c-39rh-f3gp | Medium | < 0.29.0 | 0.29.0 | 2026-09-28 |
| vLLM: Request-selected PyNvVideoCodec GPU decode bypasses static VRAM reservationCVE-2026-69147 | Medium | < 0.28.0 | 0.28.0 | 2026-09-17 |
| vLLM: Unauthenticated audio decompression-bomb DoS in /v1/chat/completionsCVE-2026-57173 | Medium | <= 0.23.0 | 0.24.0 | 2026-09-16 |
| Disaggregated generate skips decoder prompt-length validation for some multimodal processorsGHSA-3mqx-f33v-vgp9 | Medium | < 0.29.0 | 0.29.0 | 2026-09-12 |
| Incomplete artifact pin propagation in FunAudioChat and Tarsier2GHSA-hhv2-872h-628q | Medium | >= 0.22.1, < 0.28.0 | 0.28.0 | 2026-09-12 |
| Sampler Subclass Counter Shadowing Bypasses PyNvVideoCodec Decoder Limits and GPU Memory AccountingCVE-2026-100649 | Low | <0.29.0 | 0.29.0 | 2026-09-12 |
| Uncontrolled resource consumption: multimodal chat audio decoding ignores `VLLM_MAX_AUDIO_CLIP_FILESIZE_MB`CVE-2026-100648 | Medium | < 0.29.0 | 0.29.0 | 2026-09-12 |
| Remote media is fetched and fully materialized before the documented media size / item limits are enforced — remote pre-inference memory andGHSA-p6g9-7v3x-m8mv | Medium | <= 0.29.0 | No fix yet | 2026-09-12 |
| Rust HTTP/gRPC stop_token_ids bypass Python vocab-bound fix and can terminate EngineCoreGHSA-qff2-492f-9fm4 | Medium | >= 0.22.0, < 0.24.0 | 0.24.0 | 2026-09-12 |
| Out-of-range `stop_token_ids` with `min_tokens` can kill vLLM EngineCoreGHSA-v5gm-qgmv-gc6c | Medium | < 0.29.0 | 0.29.0 | 2026-09-12 |
| vLLM DoS via unbounded `cache_salt` length: multi-hundred-MB salt stalls the single EngineCore scheduler thread (CPU exhaustion)CVE-2026-100647 | Medium | <0.29.0 | 0.29.0 | 2026-09-12 |
| vLLM: SSRF + arbitrary local file read in MiMoV2OmniMultiModalProcessor `_fetch_image` and audio loader bypass MediaConnector protectionsCVE-2026-73560 | Medium | < 0.26.0 | 0.26.0 | 2026-09-08 |
| vLLM: Cross-User Data Leak VulnerabilityCVE-2026-73558 | Medium | < 0.27.0 | 0.27.0 | 2026-09-08 |
| vLLM: ReDoS via structured_outputs.regex in the lm-format-enforcer backend (no compile timeout) — missed sibling of GHSA-rwxx-mrjm-wc2mCVE-2026-73556 | Medium | < 0.26.0 | 0.26.0 | 2026-09-04 |
| vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output boundsCVE-2026-71486 | Medium | < 0.26.0 | 0.26.0 | 2026-09-04 |
| vLLM: Unauthenticated Internal Path and Username Disclosure via Validation Error MessagesCVE-2026-73555 | Medium | < 0.26.0 | 0.26.0 | 2026-09-04 |
| vLLM: Incomplete CVE-2025-62164 remediation can be bypassed by concurrent prompt partsCVE-2026-73557 | Medium | >= 0.21.0, < 0.26.0 | 0.26.0 | 2026-09-04 |
| Denial of service: a negative token id in `/v1/embeddings` or `/pooling` input kills the vLLM engine via a CUDA device-side assertionCVE-2026-93592 | High | < 0.28.0 | 0.28.0 | 2026-09-03 |
| LlavaOnevision2 processor loader executes attacker model code with `trust_remote_code=False` (inert `trust_remote_code` kwarg to `transformeCVE-2026-90553 | High | < 0.28.0 | 0.28.0 | 2026-08-28 |
| Denial of Service NanoNemoTronVL Video Audio Extraction BombCVE-2026-90554 | Medium | >=0.10.2, < 0.28.0 | 0.28.0 | 2026-08-28 |
| Speech-to-text audio decode duration limit bypass via forged header sample rateCVE-2026-90555 | Medium | < 0.28.0 | 0.28.0 | 2026-08-28 |
| vLLM: Completion prompt lists fan out into unbounded engine requestsCVE-2026-73559 | Medium | >= 0.19.0, < 0.26.0 | 0.26.0 | 2026-08-13 |
| vLLM Unauthenticated Requests Exploit DeepStream Backend Confusion for DoSCVE-2026-78684 | Medium | < 0.27.0 | 0.27.0 | 2026-08-11 |
| vLLM denial of service via prompt embeds on M-RoPE modelsCVE-2026-55514 | High | >= 0.12.0, < 0.24.0 | 0.24.0 | 2026-07-20 |
| vLLM: Processing differential in multi-channel audio downmixing enables hidden-input/moderation bypass for audio modelsCVE-2026-34760 | Medium | >= 0.5.5, < 0.18.0 | 0.18.0 | 2026-07-17 |
| vLLM has Remote DoS via Invalid Recovered Token ReinjectionCVE-2026-54234 | High | >= 0.17.1, < 0.24.0 | 0.24.0 | 2026-07-17 |
| vLLM: ReDoS via structured_outputs.regex compiled without timeout in xgrammar and outlines backendsCVE-2026-55574 | High | < 0.24.0 | 0.24.0 | 2026-07-17 |
| vLLM: Speech-to-text upload size limit is enforced after full UploadFile readCVE-2026-55646 | Medium | >= 0.22.0, < 0.24.0 | 0.24.0 | 2026-07-17 |
| vLLM: GGUF dequantize kernel int truncation exposes uninitialized GPU memory in multi-tenant servingCVE-2026-53923 | Medium | >= 0.5.5, < 0.24.0 | 0.24.0 | 2026-06-17 |
| vLLM: OOM Denial of Service via Audio Decompression BombCVE-2026-54233 | Medium | <= 0.23.0 | 0.24.0 | 2026-06-17 |
| vLLM: temperature=NaN and temperature=Infinity bypass validation and propagate to GPU kernelsCVE-2026-54235 | Medium | >= 0.8.5, <= 0.23.0 | 0.24.0 | 2026-06-17 |
| vLLM: image EXIF Rotation & PNG tRNS Transparency Not Normalized, Causing Mismatch Between Model Input and ExpectationsCVE-2026-12491 | Medium | >= 0.11.0, < 0.24.0 | 0.24.0 | 2026-06-17 |
| vLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via Anthropic routerCVE-2026-54236 | Medium | <= 0.23.0 | 0.24.0 | 2026-06-17 |
| vLLM: OpenAI auth bypassCVE-2026-48746 | Critical | >= 0.3.0, < 0.22.0 | 0.22.0 | 2026-06-16 |
| vLLM: Security Check Bypass via assert Statement in Activation Function Loading Allows Arbitrary Code ExecutionCVE-2026-41523 | High | < 0.22.0 | 0.22.0 | 2026-06-16 |
| vLLM is vulnerable to an Out-of-Memory (OOM) Denial of Service (DoS) attack due to unbounded frame count processing in the `VideoMediaIO.loaCVE-2026-5497 | High | >= 0.8.0, < 0.19.0 | 0.19.0 | 2026-06-11 |
| vLLM's Artifact Pin Decay allows pinned deployments to load unpinned code, weights, and processorsCVE-2026-47155 | Medium | < 0.22.0 | 0.22.0 | 2026-06-10 |
| vllm has Improper Resource Shutdown or ReleaseCVE-2026-9540 | Medium | <= 0.19.0 | No fix yet | 2026-05-26 |
| vLLM: extract_hidden_states speculative decoding crashes server on any request with penalty parametersCVE-2026-44223 | Medium | >= 0.18.0, < 0.20.0 | 0.20.0 | 2026-05-06 |
| vLLM Vulnerable to Remote DoS via Special-Token PlaceholdersCVE-2026-44222 | Medium | >= 0.6.1, < 0.20.0 | 0.20.0 | 2026-05-05 |
| vLLM makes Use of Uninitialized ResourceCVE-2026-7141 | Low | < 0.19.1 | 0.19.1 | 2026-04-27 |
| vLLM: Unauthenticated OOM Denial of Service via Unbounded `n` Parameter in OpenAI API ServerCVE-2026-34756 | Medium | >= 0.1.0, < 0.19.0 | 0.19.0 | 2026-04-03 |
| vLLM: Server-Side Request Forgery (SSRF) in `download_bytes_from_url `CVE-2026-34753 | Medium | >= 0.16.0, < 0.19.0 | 0.19.0 | 2026-04-03 |
| vLLM: Denial of Service via Unbounded Frame Count in video/jpeg Base64 ProcessingCVE-2026-34755 | Medium | >= 0.7.0, < 0.19.0 | 0.19.0 | 2026-04-03 |
| vLLM has Hardcoded Trust Override in Model Files Enables RCE Despite Explicit User Opt-OutCVE-2026-27893 | High | >= 0.10.1, < 0.18.0 | 0.18.0 | 2026-03-27 |
| vLLM has SSRF Protection BypassCVE-2026-25960 | Medium | >= 0.15.1, < 0.17.0 | 0.17.0 | 2026-03-09 |
| vLLM has RCE In Video ProcessingCVE-2026-22778 | Critical | >= 0.8.3, < 0.14.1 | 0.14.1 | 2026-02-02 |
| vLLM vulnerable to Server-Side Request Forgery (SSRF) through MediaConnectorCVE-2026-24779 | High | < 0.14.1 | 0.14.1 | 2026-01-28 |
| vLLM affected by RCE via auto_map dynamic module loading during model initializationCVE-2026-22807 | High | >= 0.10.1, < 0.14.0 | 0.14.0 | 2026-01-21 |
| vLLM is vulnerable to DoS in Idefics3 vision models via image payload with ambiguous dimensionsCVE-2026-22773 | Medium | >= 0.6.4, < 0.12.0 | 0.12.0 | 2026-01-13 |
| vLLM introduced enhanced protection for CVE-2025-62164CVE-2026-56340 | High | >= 0.10.2, < 0.13.0 | 0.13.0 | 2026-01-08 |
| vLLM vulnerable to remote code execution via transformers_utils/get_configCVE-2025-66448 | High | < 0.11.1 | 0.11.1 | 2025-12-02 |
| vLLM vulnerable to DoS via large Chat Completion or Tokenization requests with specially crafted `chat_template_kwargs`CVE-2025-62426 | Medium | >= 0.5.5, < 0.11.1 | 0.11.1 | 2025-11-20 |
| vLLM deserialization vulnerability leading to DoS and potential RCECVE-2025-62164 | High | >= 0.10.2, < 0.11.1 | 0.11.1 | 2025-11-20 |
| vLLM vulnerable to DoS with incorrect shape of multimodal embedding inputsCVE-2025-62372 | High | >= 0.5.5, < 0.11.1 | 0.11.1 | 2025-11-20 |
| vLLM is vulnerable to Server-Side Request Forgery (SSRF) through `MediaConnector` classCVE-2025-6242 | High | >= 0.5.0, < 0.11.0 | 0.11.0 | 2025-10-07 |
| vLLM: Resource-Exhaustion (DoS) through Malicious Jinja Template in OpenAI-Compatible ServerCVE-2025-61620 | Medium | >= 0.5.1, < 0.11.0 | 0.11.0 | 2025-10-07 |
| vLLM is vulnerable to timing attack at bearer authCVE-2025-59425 | High | < 0.11.0 | 0.11.0 | 2025-10-07 |
| vLLM has remote code execution vulnerability in the tool call parser for Qwen3-CoderCVE-2025-9141 | High | >= 0.10.0, < 0.10.1.1 | 0.10.1.1 | 2025-08-21 |
| vllm API endpoints vulnerable to Denial of Service AttacksCVE-2025-48956 | High | >= 0.1.0, < 0.10.1.1 | 0.10.1.1 | 2025-08-21 |
| Potential Timing Side-Channel Vulnerability in vLLM’s Chunk-Based Prefix CachingCVE-2025-46570 | Low | < 0.9.0 | 0.9.0 | 2025-05-28 |
| vLLM DOS: Remotely kill vllm over http with invalid JSON schemaCVE-2025-48942 | Medium | >= 0.8.0, < 0.9.0 | 0.9.0 | 2025-05-28 |
| vLLM allows clients to crash the openai server with invalid regexCVE-2025-48943 | Medium | >= 0.8.0, < 0.9.0 | 0.9.0 | 2025-05-28 |
| vLLM has a Weakness in MultiModalHasher Image Hashing ImplementationCVE-2025-46722 | Medium | >= 0.7.0, < 0.9.0 | 0.9.0 | 2025-05-28 |
| vLLM vulnerable to Regular Expression Denial of ServiceGHSA-j828-28rj-hfhp | Medium | >= 0.6.3, < 0.9.0 | 0.9.0 | 2025-05-28 |
| vLLM Tool Schema allows DoS via Malformed pattern and type FieldsCVE-2025-48944 | Medium | >= 0.8.0, < 0.9.0 | 0.9.0 | 2025-05-28 |
| vLLM has a Regular Expression Denial of Service (ReDoS, Exponential Complexity) Vulnerability in `pythonic_tool_parser.py`CVE-2025-48887 | Medium | >= 0.6.4, < 0.9.0 | 0.9.0 | 2025-05-28 |
| vLLM Allows Remote Code Execution via PyNcclPipe Communication ServiceCVE-2025-47277 | Critical | >= 0.6.5, < 0.8.5 | 0.8.5 | 2025-05-20 |
| Remote Code Execution Vulnerability in vLLM Multi-Node Cluster ConfigurationCVE-2025-30165 | High | >= 0.5.2, < 0.10.0 | 0.10.0 | 2025-05-06 |
| Data exposure via ZeroMQ on multi-node vLLM deploymentCVE-2025-30202 | High | >= 0.5.2, < 0.8.5 | 0.8.5 | 2025-04-29 |
| vLLM Vulnerable to Remote Code Execution via Mooncake IntegrationCVE-2025-32444 | Critical | >= 0.6.5, < 0.8.5 | 0.8.5 | 2025-04-29 |
| vLLM: Quadratic Time Complexity in Input Token Processing leads to denial of serviceCVE-2025-46560 | Medium | >= 0.8.0, < 0.8.5 | 0.8.5 | 2025-04-29 |
| CVE-2025-24357 Malicious model remote code execution fix bypass with PyTorch < 2.6.0GHSA-ggpf-24jw-3fcw | Critical | < 0.8.0 | 0.8.0 | 2025-04-23 |
| vLLM vulnerable to Denial of Service by abusing xgrammar cacheGHSA-hf3c-wxg2-49q9 | Medium | >= 0.6.5, < 0.8.4 | 0.8.4 | 2025-04-15 |
| vLLM Deserialization of Untrusted Data vulnerabilityCVE-2024-11041 | Critical | <= 0.6.2 | No fix yet | 2025-03-20 |
| vLLM allows Remote Code Execution by Pickle Deserialization via AsyncEngineRPCServer() RPC server entrypointsCVE-2024-9053 | Critical | <= 0.6.0 | No fix yet | 2025-03-20 |
| vLLM deserialization vulnerability in vllm.distributed.GroupCoordinator.recv_objectCVE-2024-9052 | Critical | <= 0.8.1 | No fix yet | 2025-03-20 |
| vLLM denial of service via outlines unbounded cache on diskCVE-2025-29770 | Medium | < 0.8.0 | 0.8.0 | 2025-03-19 |
| vLLM Allows Remote Code Execution via Mooncake IntegrationCVE-2025-29783 | Critical | >= 0.6.5, < 0.8.0 | 0.8.0 | 2025-03-19 |
| vLLM uses Python 3.12 built-in hash() which leads to predictable hash collisions in prefix cacheCVE-2025-25183 | Low | < 0.7.2 | 0.7.2 | 2025-02-06 |
| vllm: Malicious model to RCE by torch.load in hf_model_weights_iteratorCVE-2025-24357 | High | < 0.7.0 | 0.7.0 | 2025-01-27 |
| vLLM denial of service vulnerabilityCVE-2024-8768 | High | < 0.5.5 | 0.5.5 | 2024-09-17 |
| vLLM Denial of Service via the best_of parameterCVE-2024-8939 | Medium | <= 0.5.0.post1 | No fix yet | 2024-09-17 |