Package · PyPI
vllm
Every known advisory for the PyPI package vllm, with the affected versions and how widely it is downloaded.
Advisories194
Malicious0
In CISA KEV0
Exposure669,440 downloads last week
Newest fixed version among these advisories: d3d6bb13fb62da3234addf6574922a4ec0513d04. Check your version against each range below.
| Advisory | Affected versions | Published |
|---|---|---|
| CVE-2026-115911 GHSA-542m-9234-8gvf vLLM: DoS via unbounded chat_template Jinja2 rendering exhausts the renderer thread pool | 0.19.0 to 0.32.0 | 2026-10-28 |
| CVE-2026-115912 GHSA-6757-hggm-hp2w vLLM: Public multimodal SHM handle forgery reaches worker-side `pickle.loads()` with insecure serialization disabled | every version to 0.32.0 | 2026-10-28 |
| CVE-2026-115913 GHSA-8397-8qf6-f2gm vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engine | every version to 0.32.0 | 2026-10-28 |
| CVE-2026-115914 GHSA-gq5q-89c7-vhvx vLLM DeepSeek-OCR image processing vulnerable to denial of service | 0.11.1 to 0.32.0 | 2026-10-28 |
| CVE-2026-115915 GHSA-h3rc-j82p-2c3h vLLM: Untrusted EPS inputs can trigger denial of service or Ghostscript RCE | every version to 0.32.0 | 2026-10-28 |
| CVE-2026-115916 GHSA-q97c-p368-75c4 vLLM: Out-of-vocabulary prompt token id passes input validation and triggers a CUDA device-side assert (engine crash / DoS) | every version to 0.32.0 | 2026-10-28 |
| CVE-2026-115917 GHSA-vvpg-rrww-wpmh vLLM: Permissive default CORS origins allow cross-origin requests to the vLLM HTTP server | every version to 0.32.0 | 2026-10-28 |
| CVE-2026-115918 GHSA-x567-49g5-p4cf vLLM: Responses API bypasses trust_request_chat_template for chat_template_kwargs | 0.16.0 to 0.32.0 | 2026-10-28 |
| CVE-2026-94624 GHSA-frvg-qw88-hww9 vllm p2p never connected zmq_exhaustion | every version to 0.32.0 | 2026-10-27 |
| CVE-2026-114053 GHSA-h3rc-6mm3-gc2m vLLM: Request-controlled mm_processor_kwargs.code_revision allows remote code execution | 0.7.3 to 0.31.0 | 2026-10-27 |
| CVE-2026-115911 PYSEC-2026-4344 vLLM is an inference and serving engine for large language models. From 0.19.0 until 0.32.0, authenticated callers can supply a request chat template when no server-side --chat-template is configured, or a malicious model repository can pro | 0.19.0 to 0.32.0 | 2026-10-26 |
| CVE-2026-115912 PYSEC-2026-4345 vLLM is an inference and serving engine for large language models. Prior to 0.32.0, an authenticated caller can place forged shared-memory address and monotonic_id values in public multimodal features.kwargs_data. In scale-out deployments w | every version to 0.32.0 | 2026-10-26 |
| CVE-2026-115915 PYSEC-2026-4346 vLLM is an inference and serving engine for large language models. Prior to 0.32.0, ImageMediaIO passes attacker-controlled image bytes to Pillow through Image.open(BytesIO(data)) without restricting decoder selection, and the MIME label do | 0.30.0 to 0.32.0 | 2026-10-26 |
| CVE-2026-115916 PYSEC-2026-4347 vLLM is an inference and serving engine for large language models. Prior to 0.32.0, vllm/tokenizers/hf.py computes max_token_id with an off-by-one upper bound, allowing a pre-tokenized prompt ID equal to the vocabulary size to pass validati | every version to 0.32.0 | 2026-10-26 |
| CVE-2026-115917 PYSEC-2026-4348 vLLM is an inference and serving engine for large language models. Prior to 0.32.0, the OpenAI-compatible server configures Starlette CORSMiddleware with allowed_origins set to a wildcard by default. A malicious web page can use a victim's | every version to 0.32.0 | 2026-10-26 |
| CVE-2026-115918 PYSEC-2026-4349 vLLM is an inference and serving engine for large language models. From 0.16.0 until 0.32.0, the /v1/responses endpoint allows an authenticated caller to place an attacker-controlled template in chat_template_kwargs[chat_template] when no s | 0.16.0 to 0.32.0 | 2026-10-26 |
| CVE-2026-116173 PYSEC-2026-4350 vLLM from 0.22.0 before 0.24.0 contains a denial of service vulnerability that allows remote attackers to terminate EngineCore by bypassing logprob_token_ids limits in the Rust gRPC frontend. Attackers can send Generate or GenerateStream re | 0.22.0 to 0.24.0 | 2026-10-25 |
| CVE-2026-116176 PYSEC-2026-4351 vLLM from 0.9.2 before 0.24.0 contains a server-side request forgery vulnerability that allows API callers to force outbound connections by embedding a decode address in the X-Request-Id header. Attackers can make a P2pNcclConnector prefill | 0.9.2 to 0.24.0 | 2026-10-25 |
| CVE-2026-116177 PYSEC-2026-4352 vLLM from 0.10.2 before 0.25.0 contains an uncontrolled resource consumption vulnerability that allows unauthenticated attackers to exhaust memory because the Nemotron-VL processors globally set Image.MAX_IMAGE_PIXELS to None. Attackers can | 0.10.2 to 0.25.0 | 2026-10-25 |
| CVE-2026-116178 PYSEC-2026-4353 vLLM from 0.20.0 before 0.30.0 contains a denial of service vulnerability that allows remote attackers to crash the engine by sending inconsistent multimodal metadata to the disaggregated /inference/v1/generate endpoint. Attackers can repla | 0.20.0 to 0.30.0 | 2026-10-25 |
| CVE-2026-114055 GHSA-gx7p-2j49-hfq4 vLLM : DoS via unvalidated mm_processor_kwargs in multimodal image processors (Isaac, InternVL, H2OVL, etc.) | 0.7.3 to 0.31.0 | 2026-10-23 |
| CVE-2026-114050 GHSA-p92p-rxj5-7p2x vLLM: Forged multimodal UUID + P0/P1 processor-cache eviction drift: unauthenticated remote crash of the whole vLLM engine | every version to 0.31.0 | 2026-10-23 |
| CVE-2026-114052 PYSEC-2026-4361 vLLM after 0.24.0 and before 0.31.0 contains a memory exhaustion vulnerability in MOSS-Audio model serving that caches a processor for every distinct request-controlled mm_processor_kwargs value in an unbounded dictionary. Unauthenticated a | 0.24.0 to 0.31.0 | 2026-10-21 |
| CVE-2026-114054 PYSEC-2026-4362 vLLM before 0.31.0 contains a cache aliasing vulnerability that allows authenticated attackers to collide shared cache keys by supplying caller-controlled values such as EXIF ImageID or cache_salt. Attackers can set a UUID EXIF ImageID to r | every version to 0.31.0 | 2026-10-21 |
| CVE-2026-105756 PYSEC-2026-4306 vLLM: Loose `cache_salt` validation lets a single request kill EngineCore on LMCache-MP deployments — uncaught downstream `ValueError` denial of service | every version to 0.30.0 | 2026-10-15 |
| CVE-2026-105760 PYSEC-2026-4307 vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion | 0.23.0rc2 to 0.30.0 | 2026-10-15 |
| CVE-2026-105752 PYSEC-2026-4308 vLLM: Harmony tool continuations drop `cache_salt` — restoring a cross-tenant prefix-cache membership oracle | every version to 0.30.0 | 2026-10-15 |
| CVE-2026-105754 PYSEC-2026-4309 vLLM: Scale-out disaggregated multimodal transport trusts caller-supplied features | every version to 0.30.0 | 2026-10-15 |
| CVE-2026-105758 PYSEC-2026-4310 vLLM: Qwen2-VL / Qwen3-VL video samplers bound on request-controlled max_frames, which the num_frames ceiling does not reach | 0.24.0 to 0.30.0 | 2026-10-15 |
| CVE-2026-105752 GHSA-935w-9g4m-p28p vLLM: Harmony tool continuations drop `cache_salt` — restoring a cross-tenant prefix-cache membership oracle | every version to 0.30.0 | 2026-10-06 |
| CVE-2026-105753 GHSA-ph3r-5jfg-f84f vLLM: 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 | every version to 0.28.0 | 2026-10-06 |
| CVE-2026-105922 PYSEC-2026-4319 A security flaw has been discovered in vllm-project vLLM up to 0.31.0. This impacts the function get_token_bin_counts_and_mask of the file vllm/model_executor/layers/utils.py of the component Penalty Handler. Performing a manipulation resul | every version to 0.32.0 | 2026-10-06 |
| CVE-2026-105756 GHSA-2823-qmq8-rwvj vLLM: Loose `cache_salt` validation lets a single request kill EngineCore on LMCache-MP deployments — uncaught downstream `ValueError` denial of service | every version to 0.30.0 | 2026-10-05 |
| CVE-2026-105755 GHSA-2phq-3phc-84px vLLM: Flash late-interaction scoring caches query embeddings under a caller-controlled request id — cross-request integrity break and induced errors on `/score` and `/rerank` | every version to 0.30.0 | 2026-10-05 |
| CVE-2026-105760 GHSA-58v5-2m8f-94pr vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion | 0.23.0rc2 to 0.30.0 | 2026-10-05 |
| CVE-2026-105757 GHSA-85xf-c7hm-whqw vLLM: Structured-output request errors escape the request boundary and terminate the shared EngineCore — engine-fatal denial of service (3 sites) | every version to 0.30.0 | 2026-10-05 |
| CVE-2026-105754 GHSA-ph72-cqr5-qpp7 vLLM: Scale-out disaggregated multimodal transport trusts caller-supplied features | every version to 0.30.0 | 2026-10-05 |
| CVE-2026-105758 GHSA-x6mc-67gf-chw4 vLLM: Qwen2-VL / Qwen3-VL video samplers bound on request-controlled max_frames, which the num_frames ceiling does not reach | 0.24.0 to 0.30.0 | 2026-10-05 |
| CVE-2026-105753 PYSEC-2026-4197 vLLM is an inference and serving engine for large language models. Prior to 0.28.0, the default mirrored multimodal LRU cache can commit a media hash in the frontend sender cache during multimodal rendering and before engine admission, whil | every version to 0.28.0 | 2026-10-05 |
| CVE-2026-105755 PYSEC-2026-4198 vLLM is an inference and serving engine for large language models. Prior to 0.30.0, flash late-interaction scoring at the /score and /rerank endpoints derives each worker's query_key value from the caller-controlled X-Request-Id header. A c | every version to 0.30.0 | 2026-10-05 |
| CVE-2026-105757 PYSEC-2026-4199 vLLM is an inference and serving engine for large language models. Prior to 0.30.0, structured-output request failures can escape request-scoped validation and reach the EngineCore fatal-error path. A per-request backend mismatch can re-rai | every version to 0.30.0 | 2026-10-05 |
| CVE-2026-105759 PYSEC-2026-4325 vLLM is an inference and serving engine for large language models. Prior to 0.30.0, the Rust frontend's track_http_metrics middleware records the raw HTTP method token as a Prometheus label for requests reaching registered routes. An unauth | every version to 0.30.0 | 2026-10-05 |
| CVE-2026-69147 PYSEC-2026-4178 vLLM: Request-selected PyNvVideoCodec GPU decode bypasses static VRAM reservation | every version to 0.28.0 | 2026-10-01 |
| CVE-2026-57173 PYSEC-2026-4179 vLLM: Unauthenticated audio decompression-bomb DoS in /v1/chat/completions | every version to 0.24.0 | 2026-10-01 |
| CVE-2026-103241 PYSEC-2026-4186 A flaw has been found in vllm-project vLLM up to 0.26.0. This vulnerability affects unknown code of the file rust/src/parser/src/unified/gemma4.rs of the component Gemma4UnifiedParser. Executing a manipulation can lead to denial of service. | every version to 0.27.0 | 2026-09-30 |
| CVE-2026-100650 PYSEC-2026-4184 vLLM through 0.29.0 fetches and fully materializes remote or inline media before enforcing its documented media controls (the VLLM_MAX_AUDIO_CLIP_FILESIZE_MB compressed-audio size cap, default 25 MB, and the per-modality --limit-mm-per-prom | every version to 0.30.0 | 2026-09-26 |
| CVE-2026-100652 PYSEC-2026-4185 vLLM versions 0.22.0 through 0.23.0 fail to validate stop_token_ids against vocabulary bounds in Rust HTTP and gRPC frontends, allowing out-of-vocabulary token IDs to reach MinTokensLogitsProcessor. Attackers can submit requests with min_to | 0.22.0 to 0.24.0 | 2026-09-26 |
| CVE-2026-100653 PYSEC-2026-4324 vLLM is an inference and serving engine for large language models. In versions from 0.22.1 through 0.28.0, the operator-supplied model revision pin (--revision / --code-revision) is not propagated to several Hugging Face artifact loads for | 0.22.1 to 0.28.0 | 2026-09-26 |
| CVE-2026-94622 PYSEC-2026-4004 vLLM versions through 0.29.0 contain a denial of service vulnerability in the NIXL connector's metadata handling for prefill/decode disaggregated deployments. Attackers can send requests with incomplete kv_transfer_params dictionary entries | every version to 0.30.0 | 2026-09-21 |
| CVE-2026-94623 PYSEC-2026-4005 vLLM through 0.29.0 contains a denial of service vulnerability in the NIXL connector's prefix caching implementation that fails to properly validate block counts across multi-prompt completion requests in prefill/decode disaggregated deploy | every version to 0.30.0 | 2026-09-21 |
| CVE-2026-94624 PYSEC-2026-4006 vLLM through 0.29.0 contains a denial of service vulnerability in P2P KV offloading when OffloadingConnector is configured with TieringOffloadingSpec and a peer-to-peer secondary tier. Attackers can supply arbitrary remote host and port val | every version to 0.30.0 | 2026-09-21 |
| CVE-2026-94625 PYSEC-2026-4007 vLLM through 0.29.0 contains a resource exhaustion vulnerability in MooncakeConnector where rejected prefill requests create ownerless transfer placeholders that are never reclaimed. Attackers can send rejected requests to exhaust sender ta | every version to 0.30.0 | 2026-09-21 |
| CVE-2026-94626 PYSEC-2026-4008 vLLM through 0.29.0 fails to validate the tp_size parameter in kv_transfer_params on OpenAI-compatible completion endpoints, allowing attackers to allocate unbounded memory. Attackers can supply arbitrary tp_size values in prefill/decode di | every version to 0.30.0 | 2026-09-21 |
| CVE-2026-93989 PYSEC-2026-4000 vLLM through 0.29.0 fails to properly validate bad_words token indices against the model's generation output width in SamplingParams.update_from_tokenizer(). Attackers can supply out-of-bounds token indices that corrupt logits memory of con | every version to 0.30.0 | 2026-09-19 |
| CVE-2026-93592 PYSEC-2026-3997 vLLM versions before 0.28.0 fail to validate the lower bound of token IDs in the /v1/embeddings and /pooling endpoints, allowing unauthenticated attackers to crash the engine by submitting negative token IDs. A single request with a negativ | every version to 0.28.0 | 2026-09-18 |
| CVE-2026-93840 PYSEC-2026-3998 vLLM before 0.29.0 validates allowed_token_ids against tokenizer length instead of model output logits width in SamplingParams._validate_allowed_token_ids(). Attackers can supply token IDs above the output vocabulary that pass validation, c | every version to 0.29.0 | 2026-09-18 |
| CVE-2026-93841 PYSEC-2026-3999 vLLM through 0.29.0 contains a memory corruption vulnerability in the Triton _bincount_kernel where prompt token IDs index the penalty prompt-presence bitset without bounds checking against vocabulary size. Attackers can submit multimodal a | every version to 0.30.0 | 2026-09-18 |
| CVE-2026-69147 GHSA-8pw2-6jv3-mj5j vLLM: Request-selected PyNvVideoCodec GPU decode bypasses static VRAM reservation | every version to 0.28.0 | 2026-09-17 |
| CVE-2026-93436 PYSEC-2026-3996 vLLM through 0.29.0 fails to properly clean up decode-side metadata for rejected inference requests in prefill/decode disaggregated deployments. Remote attackers can submit requests with max_tokens=0 to exhaust decode-worker memory without | every version to 0.30.0 | 2026-09-17 |
| CVE-2026-57173 GHSA-hcwq-8wjf-3gcr vLLM: Unauthenticated audio decompression-bomb DoS in /v1/chat/completions | every version to 0.24.0 | 2026-09-16 |
| CVE-2026-90553 PYSEC-2026-3985 vLLM before 0.28.0 contains a remote code execution vulnerability in the LlavaOnevision2 processor loader that ignores the trust_remote_code parameter when loading remote processor classes. Attackers can craft a malicious model with arbitra | every version to 0.28.0 | 2026-09-12 |
| CVE-2026-73556 PYSEC-2026-3933 vLLM: ReDoS via structured_outputs.regex in the lm-format-enforcer backend (no compile timeout) — missed sibling of GHSA-rwxx-mrjm-wc2m | every version to 0.26.0 | 2026-09-10 |
| CVE-2026-73560 PYSEC-2026-3934 vLLM: SSRF + arbitrary local file read in MiMoV2OmniMultiModalProcessor `_fetch_image` and audio loader bypass MediaConnector protections | every version to 0.26.0 | 2026-09-10 |
| CVE-2026-73558 PYSEC-2026-3935 vLLM: Cross-User Data Leak Vulnerability | every version to 0.27.0 | 2026-09-10 |
| CVE-2026-71486 PYSEC-2026-3936 vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output bounds | every version to 0.26.0 | 2026-09-10 |
| CVE-2026-73555 PYSEC-2026-3937 vLLM: Unauthenticated Internal Path and Username Disclosure via Validation Error Messages | every version to 0.26.0 | 2026-09-10 |
| CVE-2026-73557 PYSEC-2026-3938 vLLM: Incomplete CVE-2025-62164 remediation can be bypassed by concurrent prompt parts | 0.21.0 to 0.26.0 | 2026-09-10 |
| CVE-2026-73560 GHSA-4hhp-h66f-j5j7 vLLM: SSRF + arbitrary local file read in MiMoV2OmniMultiModalProcessor `_fetch_image` and audio loader bypass MediaConnector protections | every version to 0.26.0 | 2026-09-08 |
| CVE-2026-73558 GHSA-7m6h-x95x-82q5 vLLM: Cross-User Data Leak Vulnerability | every version to 0.27.0 | 2026-09-08 |
| CVE-2026-73556 GHSA-48jh-3gj7-fg8v vLLM: ReDoS via structured_outputs.regex in the lm-format-enforcer backend (no compile timeout) — missed sibling of GHSA-rwxx-mrjm-wc2m | every version to 0.26.0 | 2026-09-04 |
| CVE-2026-71486 GHSA-8737-qx52-hjff vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output bounds | every version to 0.26.0 | 2026-09-04 |
| CVE-2026-73555 GHSA-hwrm-c4cx-rf4j vLLM: Unauthenticated Internal Path and Username Disclosure via Validation Error Messages | every version to 0.26.0 | 2026-09-04 |
| CVE-2026-73557 GHSA-pr7f-p5mw-fc87 vLLM: Incomplete CVE-2025-62164 remediation can be bypassed by concurrent prompt parts | 0.21.0 to 0.26.0 | 2026-09-04 |
| CVE-2026-78684 PYSEC-2026-4003 vLLM before 0.27.0 fails to properly classify DeepStream as a GPU backend and omits pixel-limit enforcement in its decode path. Unauthenticated attackers can activate DeepStream at request time to initialize the process-wide GPU decode pool | every version to 0.27.0 | 2026-08-25 |
| CVE-2026-73559 PYSEC-2026-3704 vLLM: Completion prompt lists fan out into unbounded engine requests | 0.19.0 to 0.26.0 | 2026-08-19 |
| CVE-2026-73559 GHSA-87x5-vmc3-756j vLLM: Completion prompt lists fan out into unbounded engine requests | 0.19.0 to 0.26.0 | 2026-08-13 |
| CVE-2026-54234 PYSEC-2026-3542 vLLM has Remote DoS via Invalid Recovered Token Reinjection | 0.17.1 to 0.24.0 | 2026-07-23 |
| CVE-2026-55514 GHSA-33cg-gxv8-3p8g vLLM denial of service via prompt embeds on M-RoPE models | 0.12.0 to 0.24.0 | 2026-07-20 |
| CVE-2026-34760 GHSA-6c4r-fmh3-7rh8 vLLM: Processing differential in multi-channel audio downmixing enables hidden-input/moderation bypass for audio models | 0.5.5 to 0.18.0 | 2026-07-17 |
| CVE-2026-54234 GHSA-8wr5-jm2h-8r4f vLLM has Remote DoS via Invalid Recovered Token Reinjection | 0.17.1 to 0.24.0 | 2026-07-17 |
| CVE-2026-55574 GHSA-rwxx-mrjm-wc2m vLLM: ReDoS via structured_outputs.regex compiled without timeout in xgrammar and outlines backends | every version to 0.24.0 | 2026-07-17 |
| CVE-2026-55646 GHSA-v82g-2437-67m2 vLLM: Speech-to-text upload size limit is enforced after full UploadFile read | 0.22.0 to 0.24.0 | 2026-07-17 |
| CVE-2026-53923 PYSEC-2026-3403 vLLM: GGUF dequantize kernel int truncation exposes uninitialized GPU memory in multi-tenant serving | 0.5.5 to ≤ 0.23.0 | 2026-07-13 |
| CVE-2026-54233 PYSEC-2026-3404 vLLM: OOM Denial of Service via Audio Decompression Bomb | every version to ≤ 0.23.0 | 2026-07-13 |
| CVE-2026-54235 PYSEC-2026-3405 vLLM: temperature=NaN and temperature=Infinity bypass validation and propagate to GPU kernels | every version to ≤ 0.23.0 | 2026-07-13 |
| CVE-2026-12491 PYSEC-2026-3406 vLLM: image EXIF Rotation & PNG tRNS Transparency Not Normalized, Causing Mismatch Between Model Input and Expectations | 0.11.0 to ≤ 0.23.0 | 2026-07-13 |
| CVE-2026-9540 PYSEC-2026-3407 vllm has Improper Resource Shutdown or Release | every version to ≤ 0.19.0 | 2026-07-13 |
| CVE-2026-54236 PYSEC-2026-3408 vLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via Anthropic router | every version to ≤ 0.23.0 | 2026-07-13 |
| CVE-2026-44222 PYSEC-2026-3409 vLLM Vulnerable to Remote DoS via Special-Token Placeholders | 0.6.1 to 0.20.0 | 2026-07-13 |
| CVE-2026-34753 PYSEC-2026-3410 vLLM: Server-Side Request Forgery (SSRF) in `download_bytes_from_url ` | 0.16.0 to 0.19.0 | 2026-07-13 |
| CVE-2026-25960 PYSEC-2026-3411 vLLM has SSRF Protection Bypass | 0.15.1 to 0.17.0 | 2026-07-13 |
| CVE-2026-22807 PYSEC-2026-2010 vLLM affected by RCE via auto_map dynamic module loading during model initialization | 0.10.1 to 0.14.0 | 2026-07-07 |
| CVE-2025-6242 PYSEC-2026-2011 vLLM is vulnerable to Server-Side Request Forgery (SSRF) through `MediaConnector` class | 0.5.0 to 0.11.0 | 2026-07-07 |
| CVE-2025-62426 PYSEC-2026-2012 vLLM vulnerable to DoS via large Chat Completion or Tokenization requests with specially crafted `chat_template_kwargs` | 0.5.5 to 0.11.1 | 2026-07-07 |
| CVE-2025-61620 PYSEC-2026-2013 vLLM: Resource-Exhaustion (DoS) through Malicious Jinja Template in OpenAI-Compatible Server | 0.5.1 to 0.11.0 | 2026-07-07 |
| CVE-2025-9141 PYSEC-2026-2014 vLLM has remote code execution vulnerability in the tool call parser for Qwen3-Coder | 0.10.0 to 0.10.1.1 | 2026-07-07 |
| CVE-2025-66448 PYSEC-2026-2015 vLLM vulnerable to remote code execution via transformers_utils/get_config | every version to 0.11.1 | 2026-07-07 |
| CVE-2025-30202 PYSEC-2026-2016 Data exposure via ZeroMQ on multi-node vLLM deployment | 0.5.2 to 0.8.5 | 2026-07-07 |
| CVE-2025-30165 PYSEC-2026-2017 Remote Code Execution Vulnerability in vLLM Multi-Node Cluster Configuration | 0.5.2 to 0.10.0 | 2026-07-07 |
| CVE-2025-62164 PYSEC-2026-2018 vLLM deserialization vulnerability leading to DoS and potential RCE | 0.10.2 to 0.11.1 | 2026-07-07 |
| CVE-2025-62372 PYSEC-2026-2019 vLLM vulnerable to DoS with incorrect shape of multimodal embedding inputs | 0.5.5 to 0.11.1 | 2026-07-07 |
| CVE-2026-24779 PYSEC-2026-2020 vLLM vulnerable to Server-Side Request Forgery (SSRF) through MediaConnector | every version to 0.14.1 | 2026-07-07 |
| CVE-2025-48956 PYSEC-2026-2021 vllm API endpoints vulnerable to Denial of Service Attacks | 0.1.0 to 0.10.1.1 | 2026-07-07 |
| CVE-2025-46560 PYSEC-2026-2022 phi4mm: Quadratic Time Complexity in Input Token Processing leads to denial of service | 0.8.0 to 0.8.5 | 2026-07-07 |
| CVE-2025-48944 PYSEC-2026-2023 vLLM Tool Schema allows DoS via Malformed pattern and type Fields | 0.8.0 to 0.9.0 | 2026-07-07 |
| CVE-2024-8768 PYSEC-2026-2024 vLLM denial of service vulnerability | every version to 0.5.5 | 2026-07-07 |
| CVE-2024-8939 PYSEC-2026-2025 vLLM Denial of Service via the best_of parameter | every version to ≤ 0.5.0.post1 | 2026-07-07 |
| CVE-2025-59425 PYSEC-2026-2026 vLLM is vulnerable to timing attack at bearer auth | every version to 0.11.0 | 2026-07-07 |
| CVE-2026-55514 PYSEC-2026-2303 vLLM is a library for LLM inference and serving. From 0.12.0 to before 0.24.0, sending a pure prompt embeds payload in a /v1/completions request with a model using M-RoPE causes EngineCore to fail an assertion and fatally crash, shutting do | 0.12.0 to 0.24.0 | 2026-07-06 |
| CVE-2026-55574 PYSEC-2026-2304 vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Prior to 0.24.0, the structured_outputs.regex API parameter passes a user-supplied regular expression string directly to the grammar compiler backends wit | every version to 0.24.0 | 2026-07-06 |
| CVE-2026-55646 PYSEC-2026-2305 vLLM is an inference and serving engine for large language models. From 0.22.0 to 0.23.0, the /v1/audio/transcriptions and /v1/audio/translations routes call request.file.read() to fully materialize an uploaded audio file into memory before | 0.22.0 to 0.24.0 | 2026-07-06 |
| CVE-2026-22778 PYSEC-2026-565 vLLM has RCE In Video Processing | 0.8.3 to 0.14.1 | 2026-06-29 |
| CVE-2024-11041 PYSEC-2026-566 vLLM Deserialization of Untrusted Data vulnerability | every version to ≤ 0.6.2 | 2026-06-29 |
| CVE-2025-47277 PYSEC-2026-567 vLLM Allows Remote Code Execution via PyNcclPipe Communication Service | 0.6.5 to 0.8.5 | 2026-06-29 |
| CVE-2024-9052 PYSEC-2026-568 vLLM deserialization vulnerability in vllm.distributed.GroupCoordinator.recv_object | every version to ≤ 0.8.1 | 2026-06-29 |
| CVE-2026-48746 PYSEC-2026-226 vLLM is an inference and serving engine for large language models (LLMs). From 0.3.0 until 0.22.0, a vulnerability in ASGI web servers and starlette's trust on those web servers enables an authentication bypass of the OpenAI API Authenticat | 0.3.0 to 0.22.0 | 2026-06-22 |
| CVE-2026-54232 PYSEC-2026-227 vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.1, the vLLM Dockerfile is vulnerable to a dependency confusion attack through the flashinfer-jit-cache package. The package is installed from a custom i | every version to 0.22.1 | 2026-06-22 |
| CVE-2026-41523 PYSEC-2026-2300 vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.0, an assert-based security check in vLLM's activation function loading allows any unauthenticated attacker to achieve arbitrary code execution on the s | every version to 0.22.0 | 2026-06-22 |
| CVE-2026-47155 PYSEC-2026-2301 vLLM is an inference and serving engine for large language models (LLMs). Prior to 0.22.0, vLLM's revision pinning controls do not consistently apply to all artifacts loaded for a model. A deployment that supplies --revision or --code-revis | every version to 0.22.0 | 2026-06-22 |
| CVE-2026-56340 PYSEC-2026-250 vLLM versions >= 0.10.2 and < 0.13.0 are missing sparse tensor validation in multimodal embeddings processing. Because PyTorch disables sparse tensor invariant checks by default, an attacker can submit crafted embedding requests with malfor | 0.10.2 to 0.13.0 | 2026-06-20 |
| CVE-2026-53923 GHSA-5jv2-g5wq-cmr4 vLLM: GGUF dequantize kernel int truncation exposes uninitialized GPU memory in multi-tenant serving | 0.5.5 to 0.24.0 | 2026-06-17 |
| CVE-2026-54233 GHSA-6pr9-rp53-2pmc vLLM: OOM Denial of Service via Audio Decompression Bomb | every version to 0.24.0 | 2026-06-17 |
| CVE-2026-54235 GHSA-7h4p-rffg-7823 vLLM: temperature=NaN and temperature=Infinity bypass validation and propagate to GPU kernels | 0.8.5 to 0.24.0 | 2026-06-17 |
| CVE-2026-12491 GHSA-8jr5-v98p-w75m vLLM: image EXIF Rotation & PNG tRNS Transparency Not Normalized, Causing Mismatch Between Model Input and Expectations | 0.11.0 to 0.24.0 | 2026-06-17 |
| CVE-2026-54236 GHSA-hgg8-fqqc-vfmw vLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via Anthropic router | every version to 0.24.0 | 2026-06-17 |
| CVE-2026-48746 GHSA-94f4-hr76-p5j6 vLLM: OpenAI auth bypass | 0.3.0 to 0.22.0 | 2026-06-16 |
| CVE-2026-41523 GHSA-q8gq-377p-jq3r vLLM: Security Check Bypass via assert Statement in Activation Function Loading Allows Arbitrary Code Execution | every version to 0.22.0 | 2026-06-16 |
| CVE-2026-5497 GHSA-wcwg-c5fc-9vrc vLLM is vulnerable to an Out-of-Memory (OOM) Denial of Service (DoS) attack due to unbounded frame count processing in the `VideoMediaIO.load_base64()` method | 0.8.0 to 0.19.0 | 2026-06-11 |
| CVE-2026-5497 PYSEC-2026-2302 vLLM versions 0.8.0 and later are vulnerable to an Out-of-Memory (OOM) Denial of Service (DoS) attack due to unbounded frame count processing in the `VideoMediaIO.load_base64()` method. When processing `video/jpeg` data URLs, the method spl | 0.8.0 to 0.19.0 | 2026-06-11 |
| CVE-2026-47155 GHSA-3ww4-5jv9-j5gm vLLM's Artifact Pin Decay allows pinned deployments to load unpinned code, weights, and processors | every version to 0.22.0 | 2026-06-10 |
| CVE-2026-9540 GHSA-98f3-hwg4-4rf7 vllm has Improper Resource Shutdown or Release | every version to ≤ 0.19.0 | 2026-05-26 |
| CVE-2026-44223 PYSEC-2026-145 vLLM is an inference and serving engine for large language models (LLMs). From to before 0.20.0, the extract_hidden_states speculative decoding proposer in vLLM returns a tensor with an incorrect shape after the first decode step, causing | 0.18.0 to 0.20.0 | 2026-05-12 |
| CVE-2026-44223 GHSA-83vm-p52w-f9pw vLLM: extract_hidden_states speculative decoding crashes server on any request with penalty parameters | 0.18.0 to 0.20.0 | 2026-05-06 |
| CVE-2026-44222 GHSA-hpv8-x276-m59f vLLM Vulnerable to Remote DoS via Special-Token Placeholders | 0.6.1 to 0.20.0 | 2026-05-05 |
| CVE-2026-7141 GHSA-x368-4g9h-fvv4 vLLM makes Use of Uninitialized Resource | every version to 0.19.1 | 2026-04-27 |
| CVE-2026-7141 PYSEC-2026-2306 A vulnerability was found in vllm up to 0.19.0. The affected element is the function has_mamba_layers of the file vllm/v1/kv_cache_interface.py of the component KV Block Handler. Performing a manipulation results in uninitialized resource. | every version to 0.19.1 | 2026-04-27 |
| CVE-2026-34755 PYSEC-2026-144 vLLM is an inference and serving engine for large language models (LLMs). From 0.7.0 to before 0.19.0, the VideoMediaIO.load_base64() method at vllm/multimodal/media/video.py splits video/jpeg data URLs by comma to extract individual JPEG f | 0.7.0 to 0.19.0 | 2026-04-06 |
| CVE-2026-34756 PYSEC-2026-2298 vLLM is an inference and serving engine for large language models (LLMs). From 0.1.0 to before 0.19.0, a Denial of Service vulnerability exists in the vLLM OpenAI-compatible API server. Due to the lack of an upper bound validation on the n | 0.1.0 to 0.19.0 | 2026-04-06 |
| CVE-2026-34756 GHSA-3mwp-wvh9-7528 vLLM: Unauthenticated OOM Denial of Service via Unbounded `n` Parameter in OpenAI API Server | 0.1.0 to 0.19.0 | 2026-04-03 |
| CVE-2026-34753 GHSA-pf3h-qjgv-vcpr vLLM: Server-Side Request Forgery (SSRF) in `download_bytes_from_url ` | 0.16.0 to 0.19.0 | 2026-04-03 |
| CVE-2026-34755 GHSA-pq5c-rjhq-qp7p vLLM: Denial of Service via Unbounded Frame Count in video/jpeg Base64 Processing | 0.7.0 to 0.19.0 | 2026-04-03 |
| CVE-2026-34760 PYSEC-2026-2299 vLLM is an inference and serving engine for large language models (LLMs). From version 0.5.5 to before version 0.18.0, Librosa defaults to using numpy.mean for mono downmixing (to_mono), while the international standard ITU-R BS.775-4 speci | 0.5.5 to 0.18.0 | 2026-04-02 |
| CVE-2026-27893 GHSA-7972-pg2x-xr59 vLLM has Hardcoded Trust Override in Model Files Enables RCE Despite Explicit User Opt-Out | 0.10.1 to 0.18.0 | 2026-03-27 |
| CVE-2026-27893 PYSEC-2026-2297 vLLM is an inference and serving engine for large language models (LLMs). Starting in version 0.10.1 and prior to version 0.18.0, two model implementation files hardcode `trust_remote_code=True` when loading sub-components, bypassing the us | 0.10.1 to 0.18.0 | 2026-03-27 |
| CVE-2026-25960 GHSA-v359-jj2v-j536 vLLM has SSRF Protection Bypass | 0.15.1 to 0.17.0 | 2026-03-09 |
| CVE-2026-22778 GHSA-4r2x-xpjr-7cvv vLLM has RCE In Video Processing | 0.8.3 to 0.14.1 | 2026-02-02 |
| CVE-2026-24779 GHSA-qh4c-xf7m-gxfc vLLM vulnerable to Server-Side Request Forgery (SSRF) through MediaConnector | every version to 0.14.1 | 2026-01-28 |
| CVE-2026-22807 GHSA-2pc9-4j83-qjmr vLLM affected by RCE via auto_map dynamic module loading during model initialization | 0.10.1 to 0.14.0 | 2026-01-21 |
| CVE-2026-22773 GHSA-grg2-63fw-f2qr vLLM is vulnerable to DoS in Idefics3 vision models via image payload with ambiguous dimensions | 0.6.4 to 0.12.0 | 2026-01-13 |
| CVE-2026-22773 PYSEC-2026-143 vLLM is an inference and serving engine for large language models (LLMs). In versions from 0.6.4 to before 0.12.0, users can crash the vLLM engine serving multimodal models that use the Idefics3 vision model implementation by sending a spec | 0.6.4 to 0.12.0 | 2026-01-10 |
| CVE-2026-56340 GHSA-mcmc-2m55-j8jj vLLM introduced enhanced protection for CVE-2025-62164 | 0.10.2 to 0.13.0 | 2026-01-08 |
| CVE-2025-66448 GHSA-8fr4-5q9j-m8gm vLLM vulnerable to remote code execution via transformers_utils/get_config | every version to 0.11.1 | 2025-12-02 |
| CVE-2025-62426 GHSA-69j4-grxj-j64p vLLM vulnerable to DoS via large Chat Completion or Tokenization requests with specially crafted `chat_template_kwargs` | 0.5.5 to 0.11.1 | 2025-11-20 |
| CVE-2025-62164 GHSA-mrw7-hf4f-83pf vLLM deserialization vulnerability leading to DoS and potential RCE | 0.10.2 to 0.11.1 | 2025-11-20 |
| CVE-2025-62372 GHSA-pmqf-x6x8-p7qw vLLM vulnerable to DoS with incorrect shape of multimodal embedding inputs | 0.5.5 to 0.11.1 | 2025-11-20 |
| CVE-2025-6242 GHSA-3f6c-7fw2-ppm4 vLLM is vulnerable to Server-Side Request Forgery (SSRF) through `MediaConnector` class | 0.5.0 to 0.11.0 | 2025-10-07 |
| CVE-2025-61620 GHSA-6fvq-23cw-5628 vLLM: Resource-Exhaustion (DoS) through Malicious Jinja Template in OpenAI-Compatible Server | 0.5.1 to 0.11.0 | 2025-10-07 |
| CVE-2025-59425 GHSA-wr9h-g72x-mwhm vLLM is vulnerable to timing attack at bearer auth | every version to 0.11.0 | 2025-10-07 |
| CVE-2025-9141 GHSA-79j6-g2m3-jgfw vLLM has remote code execution vulnerability in the tool call parser for Qwen3-Coder | 0.10.0 to 0.10.1.1 | 2025-08-21 |
| CVE-2025-48956 GHSA-rxc4-3w6r-4v47 vllm API endpoints vulnerable to Denial of Service Attacks | 0.1.0 to 0.10.1.1 | 2025-08-21 |
| CVE-2025-48887 PYSEC-2025-50 vLLM, an inference and serving engine for large language models (LLMs), has a Regular Expression Denial of Service (ReDoS) vulnerability in the file `vllm/entrypoints/openai/tool_parsers/pythonic_tool_parser.py` of versions 0.6.4 up to but | every version to 4fc1bf813ad80172c1db31264beaef7d93fe0601 0.6.4 to 0.9.0 | 2025-05-30 |
| CVE-2025-48942 PYSEC-2025-54 vLLM is an inference and serving engine for large language models (LLMs). In versions 0.8.0 up to but excluding 0.9.0, hitting the /v1/completions API with a invalid json_schema as a Guided Param kills the vllm server. This vulnerability i | every version to 08bf7840780980c7568c573c70a6a8db94fd45ff 0.8.0 to 0.9.0 | 2025-05-30 |
| CVE-2025-48943 PYSEC-2025-55 vLLM is an inference and serving engine for large language models (LLMs). Version 0.8.0 up to but excluding 0.9.0 have a Denial of Service (ReDoS) that causes the vLLM server to crash if an invalid regex was provided while using structured | every version to 08bf7840780980c7568c573c70a6a8db94fd45ff 0.8.0 to 0.9.0 | 2025-05-30 |
| CVE-2025-46722 PYSEC-2025-43 vLLM is an inference and serving engine for large language models (LLMs). In versions starting from 0.7.0 to before 0.9.0, in the file vllm/multimodal/hasher.py, the MultiModalHasher class has a security and data integrity issue in its imag | every version to 99404f53c72965b41558aceb1bc2380875f5d848 every version to 0.9.0 | 2025-05-29 |
| CVE-2025-46570 PYSEC-2025-53 vLLM is an inference and serving engine for large language models (LLMs). Prior to version 0.9.0, when a new prompt is processed, if the PageAttention mechanism finds a matching prefix chunk, the prefill process speeds up, which is reflecte | every version to 77073c77bc2006eb80ea6d5128f076f5e6c6f54f every version to 0.9.0 | 2025-05-29 |
| CVE-2025-46570 GHSA-4qjh-9fv9-r85r Potential Timing Side-Channel Vulnerability in vLLM’s Chunk-Based Prefix Caching | every version to 0.9.0 | 2025-05-28 |
| CVE-2025-48942 GHSA-6qc9-v4r8-22xg vLLM DOS: Remotely kill vllm over http with invalid JSON schema | 0.8.0 to 0.9.0 | 2025-05-28 |
| CVE-2025-48943 GHSA-9hcf-v7m4-6m2j vLLM allows clients to crash the openai server with invalid regex | 0.8.0 to 0.9.0 | 2025-05-28 |
| CVE-2025-46722 GHSA-c65p-x677-fgj6 vLLM has a Weakness in MultiModalHasher Image Hashing Implementation | 0.7.0 to 0.9.0 | 2025-05-28 |
| CVE-2025-71379 GHSA-j828-28rj-hfhp vLLM vulnerable to Regular Expression Denial of Service | 0.6.3 to 0.9.0 | 2025-05-28 |
| CVE-2025-48944 GHSA-vrq3-r879-7m65 vLLM Tool Schema allows DoS via Malformed pattern and type Fields | 0.8.0 to 0.9.0 | 2025-05-28 |
| CVE-2025-48887 GHSA-w6q7-j642-7c25 vLLM has a Regular Expression Denial of Service (ReDoS, Exponential Complexity) Vulnerability in `pythonic_tool_parser.py` | 0.6.4 to 0.9.0 | 2025-05-28 |
| CVE-2025-47277 GHSA-hjq4-87xh-g4fv vLLM Allows Remote Code Execution via PyNcclPipe Communication Service | 0.6.5 to 0.8.5 | 2025-05-20 |
| CVE-2025-30165 GHSA-9pcc-gvx5-r5wm Remote Code Execution Vulnerability in vLLM Multi-Node Cluster Configuration | 0.5.2 to 0.10.0 | 2025-05-06 |
| CVE-2025-32444 PYSEC-2025-42 vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Versions starting from 0.6.5 and prior to 0.8.5, having vLLM integration with mooncake, are vulnerable to remote code execution due to using pickle based | every version to a5450f11c95847cf51a17207af9a3ca5ab569b2c 0.6.5 to 0.8.5 | 2025-04-30 |
| CVE-2025-30202 GHSA-9f8f-2vmf-885j Data exposure via ZeroMQ on multi-node vLLM deployment | 0.5.2 to 0.8.5 | 2025-04-29 |
| CVE-2025-32444 GHSA-hj4w-hm2g-p6w5 vLLM Vulnerable to Remote Code Execution via Mooncake Integration | 0.6.5 to 0.8.5 | 2025-04-29 |
| CVE-2025-46560 GHSA-vc6m-hm49-g9qg vLLM: Quadratic Time Complexity in Input Token Processing leads to denial of service | 0.8.0 to 0.8.5 | 2025-04-29 |
| GHSA-ggpf-24jw-3fcw CVE-2025-24357 Malicious model remote code execution fix bypass with PyTorch < 2.6.0 | every version to 0.8.0 | 2025-04-23 |
| GHSA-hf3c-wxg2-49q9 vLLM vulnerable to Denial of Service by abusing xgrammar cache | 0.6.5 to 0.8.4 | 2025-04-15 |
| CVE-2024-11041 GHSA-5vqr-wprc-cpp7 vLLM Deserialization of Untrusted Data vulnerability | every version to ≤ 0.6.2 | 2025-03-20 |
| CVE-2024-9053 GHSA-cj47-qj6g-x7r4 vLLM allows Remote Code Execution by Pickle Deserialization via AsyncEngineRPCServer() RPC server entrypoints | every version to ≤ 0.6.0 | 2025-03-20 |
| CVE-2024-9052 GHSA-pgr7-mhp5-fgjp vLLM deserialization vulnerability in vllm.distributed.GroupCoordinator.recv_object | every version to ≤ 0.8.1 | 2025-03-20 |
| CVE-2024-9053 PYSEC-2025-222 vllm-project vllm version 0.6.0 contains a vulnerability in the AsyncEngineRPCServer() RPC server entrypoints. The core functionality run_server_loop() calls the function _make_handler_coro(), which directly uses cloudpickle.loads() on rece | every version to ≤ 0.6.0 | 2025-03-20 |
| CVE-2025-29770 GHSA-mgrm-fgjv-mhv8 vLLM denial of service via outlines unbounded cache on disk | every version to 0.8.0 | 2025-03-19 |
| CVE-2025-29783 GHSA-x3m8-f7g5-qhm7 vLLM Allows Remote Code Execution via Mooncake Integration | 0.6.5 to 0.8.0 | 2025-03-19 |
| CVE-2025-29770 PYSEC-2025-223 vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. The outlines library is one of the backends used by vLLM to support structured output (a.k.a. guided decoding). Outlines provides an optional cache for it | every version to 0.8.0 | 2025-03-19 |
| CVE-2025-29783 PYSEC-2025-63 vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. When vLLM is configured to use Mooncake, unsafe deserialization exposed directly over ZMQ/TCP on all network interfaces will allow attackers to execute re | every version to 288ca110f68d23909728627d3100e5a8db820aa2 0.6.5 to 0.8.0 | 2025-03-19 |
| CVE-2025-25183 PYSEC-2025-62 vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs. Maliciously constructed statements can lead to hash collisions, resulting in cache reuse, which can interfere with subsequent responses and cause unintend | every version to 432117cd1f59c76d97da2eaff55a7d758301dbc7 every version to 0.7.2 | 2025-02-07 |
| CVE-2025-25183 GHSA-rm76-4mrf-v9r8 vLLM uses Python 3.12 built-in hash() which leads to predictable hash collisions in prefix cache | every version to 0.7.2 | 2025-02-06 |
| CVE-2025-24357 GHSA-rh4j-5rhw-hr54 vllm: Malicious model to RCE by torch.load in hf_model_weights_iterator | every version to 0.7.0 | 2025-01-27 |
| CVE-2025-24357 PYSEC-2025-58 vLLM is a library for LLM inference and serving. vllm/model_executor/weight_utils.py implements hf_model_weights_iterator to load the model checkpoint, which is downloaded from huggingface. It uses the torch.load function and the weights_on | every version to d3d6bb13fb62da3234addf6574922a4ec0513d04 every version to 0.7.0 | 2025-01-27 |
| CVE-2024-8768 GHSA-w2r7-9579-27hf vLLM denial of service vulnerability | every version to 0.5.5 | 2024-09-17 |
| CVE-2024-8939 GHSA-wc36-9694-f9rf vLLM Denial of Service via the best_of parameter | every version to ≤ 0.5.0.post1 | 2024-09-17 |
Advisories from OSV (GitHub, PyPA, RustSec, Go, OpenSSF malicious packages and more). Scan a whole lockfile on the scan page. Most exposed packages.