© 2025-2026 PySpect
First version without a known vulnerability: 0.32.0
vLLM: Permissive default CORS origins allow cross-origin requests to the vLLM HTTP server
Fixed in: 0.32.0
vLLM: Phi-4 multimodal audio encoder materializes and retains full attention matrices: a single max-length audio OOM-crashes the vLLM engine
Fixed in: 0.32.0
vLLM DeepSeek-OCR image processing vulnerable to denial of service
Fixed in: 0.32.0
vLLM: Untrusted EPS inputs can trigger denial of service or Ghostscript RCE
Fixed in: 0.32.0
vLLM: Public multimodal SHM handle forgery reaches worker-side `pickle.loads()` with insecure serialization disabled
Fixed in: 0.32.0
vLLM: Out-of-vocabulary prompt token id passes input validation and triggers a CUDA device-side assert (engine crash / DoS)
Fixed in: 0.32.0
vLLM: Responses API bypasses trust_request_chat_template for chat_template_kwargs
Fixed in: 0.32.0
vLLM: DoS via unbounded chat_template Jinja2 rendering exhausts the renderer thread pool
Fixed in: 0.32.0
vLLM: Request-controlled mm_processor_kwargs.code_revision allows remote code execution
Fixed in: 0.31.0
vllm p2p never connected zmq_exhaustion
Fixed in: 0.32.0, 0.30.0
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 replay a valid multimodal item with repeated or overlapping mm_placeholders ranges or mismatched kwargs_data and mm_hashes lists, producing a negative text span in Qwen2.5-VL position construction that terminates the engine.
Fixed in: 0.30.0
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 submit a pathological decompression-bomb image to /v1/chat/completions to OOM-kill the worker and deny service to co-tenants.
Fixed in: 0.25.0
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 worker ZMQ-connect to an attacker host, leaking the NCCL unique_id and KV-cache tensors or blocking the worker.
Fixed in: 0.24.0
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 requests with output_logprobs enabled and more than 128 output_candidates.token_ids, triggering a fatal sampler exception that crashes the engine.
Fixed in: 0.24.0
vLLM : DoS via unvalidated mm_processor_kwargs in multimodal image processors (Isaac, InternVL, H2OVL, etc.)
Fixed in: 0.31.0
vLLM: Forged multimodal UUID + P0/P1 processor-cache eviction drift: unauthenticated remote crash of the whole vLLM engine
Fixed in: 0.31.0
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 receive another request's cached processed media, or match cache_salt to a LoRA name to reuse first-block KV state.
Fixed in: 0.31.0
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 attackers can send distinct or oversized mel_config values to the /tokenize endpoint, which bypasses --api-key, to exhaust host memory and crash the API process.
Fixed in: 0.31.0
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 results in denial of service. Remote exploitation of the attack is possible. The exploit has been released to the public and may be used for attacks. The project was informed of the problem early through an issue report but has not responded yet.
Fixed in: 0.32.0
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
Fixed in: 0.28.0
vLLM: Harmony tool continuations drop `cache_salt` — restoring a cross-tenant prefix-cache membership oracle
Fixed in: 0.30.0
vLLM: Qwen2-VL / Qwen3-VL video samplers bound on request-controlled max_frames, which the num_frames ceiling does not reach
Fixed in: 0.30.0
vLLM: GLMGA video sampling permits request-driven CPU and memory exhaustion
Fixed in: 0.30.0
vLLM: Scale-out disaggregated multimodal transport trusts caller-supplied features
Fixed in: 0.30.0
vLLM: Structured-output request errors escape the request boundary and terminate the shared EngineCore — engine-fatal denial of service (3 sites)
Fixed in: 0.30.0
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`
Fixed in: 0.30.0
vLLM: Loose `cache_salt` validation lets a single request kill EngineCore on LMCache-MP deployments — uncaught downstream `ValueError` denial of service
Fixed in: 0.30.0
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 unauthenticated attacker can send unique arbitrary method tokens to unguarded routes such as /tokenize, causing Prometheus's Family::get_or_create function to permanently create counter and histogram label sets. Those label sets increase process memory usage and enlarge the /metrics response until the service or monitoring path is exhausted. This issue is fixed in version 0.30.0.
Fixed in: 0.30.0
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. The attack may be launched remotely. The exploit has been published and may be used. Upgrading to version 0.29.1rc0 is able to resolve this issue. This patch is called 3439bad37e68ba9755a46f4f6b44a4aeaf1f60a9. Upgrading the affected component is advised.
Fixed in: 0.27.0
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 the FunAudioChat and Tarsier2 architectures: the WhisperFeatureExtractor and speech_tokenizer PreTrainedTokenizerFast loads in vllm/model_executor/models/funaudiochat.py and the Qwen2VLConfig.from_pretrained call used by Tarsier2ProcessingInfo in vllm/model_executor/models/qwen2_vl.py. As a result, deployments pinned to a reviewed revision still resolve these behavior-affecting processor, tokenizer, and config artifacts from the repository's default revision, so a later change to the upstream default branch can alter audio preprocessing, speech tokenizer behavior, or Tarsier2 configuration without any change to the operator's configured pin. This is a supply-chain integrity and reproducibility failure for pinned deployments; it is residual to the earlier fix tracked as GHSA-3ww4-5jv9-j5gm / CVE-2026-47155 and does not constitute remote code execution or a trust_remote_code=False bypass. The issue is fixed in version 0.28.0.
Fixed in: 0.28.0
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_tokens greater than zero and out-of-vocabulary stop_token_ids to trigger CUDA tensor indexing failures that leave EngineCore in a fatal state requiring service restart.
Fixed in: 0.24.0
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-prompt item limits). Across four ingress paths — the shared media-acquisition layer (HTTPConnection.get_bytes()/async_get_bytes()), the chat completions audio_url/base64 path, the batch speech runner, and the Rust frontend POST /tokenize route — the server reads the entire HTTP response body, base64-decodes the inline payload, or spawns one fetch/decode task per media part, and only then applies the limit (or, on some paths, never applies it). A remote attacker can therefore cause the API server or batch-runner process to allocate memory and consume outbound bandwidth proportional to an attacker-chosen body size or media item count before the request is rejected, resulting in pre-inference memory and bandwidth exhaustion (denial of service). The chat and batch surfaces require an API key when one is configured; the Rust frontend /tokenize route is unauthenticated by design. There is no code execution or data disclosure impact.
Fixed in: 0.30.0
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 disaggregated deployments to exhaust memory and trigger kernel OOM-kill of the decode worker process.
Fixed in: 0.30.0
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 task pools, causing valid requests to be delayed by up to 480 seconds while health checks continue returning success.
Fixed in: 0.30.0
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 deployments. Attackers can trigger an assertion failure in NixlBaseConnectorWorker._apply_prefix_caching by submitting completion requests with multiple prompts of varying lengths, causing the decode worker to terminate and become unavailable until restarted.
Fixed in: 0.30.0
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 to trigger an uncaught KeyError in EngineCore scheduling, causing the decode engine to terminate and making all routed requests fail until manual restart.
Fixed in: 0.30.0
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 concurrent requests, causing different in-flight HTTP requests to return incorrect tokens.
Fixed in: 0.30.0
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 audio requests with tokens equal to vocabulary size, causing out-of-bounds writes that corrupt concurrent requests' sampler state and alter repetition penalty behavior.
Fixed in: 0.30.0
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, causing LogitBiasState to corrupt GPU logits state and allow concurrent requests to sample tokens outside their allowlists.
Fixed in: 0.29.0
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 negative token ID triggers a CUDA device-side assertion that poisons the GPU context, causing all subsequent requests to fail until the process restarts.
Fixed in: 0.28.0
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 bound until the worker restarts.
Fixed in: 0.30.0
vLLM: Request-selected PyNvVideoCodec GPU decode bypasses static VRAM reservation
Fixed in: 0.28.0
vLLM: Unauthenticated audio decompression-bomb DoS in /v1/chat/completions
Fixed in: 0.24.0
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 arbitrary code in processing_llava_onevision2.py that executes with vLLM process authority even when trust_remote_code is set to False.
Fixed in: 0.28.0
vLLM: SSRF + arbitrary local file read in MiMoV2OmniMultiModalProcessor `_fetch_image` and audio loader bypass MediaConnector protections
Fixed in: 0.26.0
vLLM: Cross-User Data Leak Vulnerability
Fixed in: 0.27.0
vLLM: Incomplete CVE-2025-62164 remediation can be bypassed by concurrent prompt parts
Fixed in: 0.26.0
vLLM: ReDoS via structured_outputs.regex in the lm-format-enforcer backend (no compile timeout) — missed sibling of GHSA-rwxx-mrjm-wc2m
Fixed in: 0.26.0
vLLM: Unauthenticated Internal Path and Username Disclosure via Validation Error Messages
Fixed in: 0.26.0
vLLM: Derender endpoints decode caller-supplied GenerateResponse token IDs without output bounds
Fixed in: 0.26.0
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 and submit video that bypasses resource controls, causing partial denial of service for concurrent requests.
Fixed in: 0.27.0
vLLM: Completion prompt lists fan out into unbounded engine requests
Fixed in: 0.26.0
vLLM denial of service via prompt embeds on M-RoPE models
Fixed in: 0.24.0
vLLM: Speech-to-text upload size limit is enforced after full UploadFile read
Fixed in: 0.24.0
vLLM: ReDoS via structured_outputs.regex compiled without timeout in xgrammar and outlines backends
Fixed in: 0.24.0
vLLM has Remote DoS via Invalid Recovered Token Reinjection
Fixed in: 0.24.0
vLLM: Processing differential in multi-channel audio downmixing enables hidden-input/moderation bypass for audio models
Fixed in: 0.18.0
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 index (flashinfer.ai/whl/) using --extra-index-url, but the package name was not registered on PyPI, and UV_INDEX_STRATEGY="unsafe-best-match" is set globally. An attacker who registers flashinfer-jit-cache on PyPI with version 0.6.11.post2 can execute arbitrary code as root during the Docker build and backdoor every resulting container image, enabling exfiltration of all user prompts, API credentials, and model data from production vLLM deployments This vulnerability is fixed in 0.22.1.
Fixed in: 0.22.1
vLLM: OOM Denial of Service via Audio Decompression Bomb
Fixed in: 0.24.0
vLLM: incomplete CVE-2026-22778 fix leaks PIL repr addresses via Anthropic router
Fixed in: 0.24.0
vLLM: GGUF dequantize kernel int truncation exposes uninitialized GPU memory in multi-tenant serving
Fixed in: 0.24.0
vLLM: image EXIF Rotation & PNG tRNS Transparency Not Normalized, Causing Mismatch Between Model Input and Expectations
Fixed in: 0.24.0
vLLM: temperature=NaN and temperature=Infinity bypass validation and propagate to GPU kernels
Fixed in: 0.24.0
vLLM: OpenAI auth bypass
Fixed in: 0.22.0
vLLM: Security Check Bypass via assert Statement in Activation Function Loading Allows Arbitrary Code Execution
Fixed in: 0.22.0
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
Fixed in: 0.19.0
vLLM's Artifact Pin Decay allows pinned deployments to load unpinned code, weights, and processors
Fixed in: 0.22.0
vllm has Improper Resource Shutdown or Release
vLLM: extract_hidden_states speculative decoding crashes server on any request with penalty parameters
Fixed in: 0.20.0
vLLM Vulnerable to Remote DoS via Special-Token Placeholders
Fixed in: 0.20.0
vLLM makes Use of Uninitialized Resource
Fixed in: 0.19.1
vLLM: Denial of Service via Unbounded Frame Count in video/jpeg Base64 Processing
Fixed in: 0.19.0
vLLM: Server-Side Request Forgery (SSRF) in `download_bytes_from_url `
Fixed in: 0.19.0
vLLM: Unauthenticated OOM Denial of Service via Unbounded `n` Parameter in OpenAI API Server
Fixed in: 0.19.0
vLLM has Hardcoded Trust Override in Model Files Enables RCE Despite Explicit User Opt-Out
Fixed in: 0.18.0
vLLM has SSRF Protection Bypass
Fixed in: 0.17.0
vLLM has RCE In Video Processing
Fixed in: 0.14.1
vLLM vulnerable to Server-Side Request Forgery (SSRF) through MediaConnector
Fixed in: 0.14.1
vLLM affected by RCE via auto_map dynamic module loading during model initialization
Fixed in: 0.14.0
vLLM is vulnerable to DoS in Idefics3 vision models via image payload with ambiguous dimensions
Fixed in: 0.12.0
vLLM introduced enhanced protection for CVE-2025-62164
Fixed in: 0.13.0
vLLM vulnerable to remote code execution via transformers_utils/get_config
Fixed in: 0.11.1
vLLM vulnerable to DoS via large Chat Completion or Tokenization requests with specially crafted `chat_template_kwargs`
Fixed in: 0.11.1
vLLM vulnerable to DoS with incorrect shape of multimodal embedding inputs
Fixed in: 0.11.1
vLLM deserialization vulnerability leading to DoS and potential RCE
Fixed in: 0.11.1
vLLM is vulnerable to Server-Side Request Forgery (SSRF) through `MediaConnector` class
Fixed in: 0.11.0
vLLM: Resource-Exhaustion (DoS) through Malicious Jinja Template in OpenAI-Compatible Server
Fixed in: 0.11.0
vLLM is vulnerable to timing attack at bearer auth
Fixed in: 0.11.0
vLLM has remote code execution vulnerability in the tool call parser for Qwen3-Coder
Fixed in: 0.10.1.1
vllm API endpoints vulnerable to Denial of Service Attacks
Fixed in: 0.10.1.1
vLLM Tool Schema allows DoS via Malformed pattern and type Fields
Fixed in: 0.9.0
vLLM allows clients to crash the openai server with invalid regex
Fixed in: 0.9.0, 08bf7840780980c7568c573c70a6a8db94fd45ff
vLLM DOS: Remotely kill vllm over http with invalid JSON schema
Fixed in: 0.9.0, 08bf7840780980c7568c573c70a6a8db94fd45ff
vLLM has a Weakness in MultiModalHasher Image Hashing Implementation
Fixed in: 0.9.0, 99404f53c72965b41558aceb1bc2380875f5d848
Potential Timing Side-Channel Vulnerability in vLLM’s Chunk-Based Prefix Caching
Fixed in: 0.9.0, 77073c77bc2006eb80ea6d5128f076f5e6c6f54f
vLLM vulnerable to Regular Expression Denial of Service
Fixed in: 0.9.0
vLLM has a Regular Expression Denial of Service (ReDoS, Exponential Complexity) Vulnerability in `pythonic_tool_parser.py`
Fixed in: 0.9.0, 4fc1bf813ad80172c1db31264beaef7d93fe0601
vLLM Allows Remote Code Execution via PyNcclPipe Communication Service
Fixed in: 0.8.5
Remote Code Execution Vulnerability in vLLM Multi-Node Cluster Configuration
Fixed in: 0.10.0
vLLM: Quadratic Time Complexity in Input Token Processing leads to denial of service
Fixed in: 0.8.5
vLLM Vulnerable to Remote Code Execution via Mooncake Integration
Fixed in: 0.8.5, a5450f11c95847cf51a17207af9a3ca5ab569b2c
Data exposure via ZeroMQ on multi-node vLLM deployment
Fixed in: 0.8.5
CVE-2025-24357 Malicious model remote code execution fix bypass with PyTorch < 2.6.0
Fixed in: 0.8.0
vLLM vulnerable to Denial of Service by abusing xgrammar cache
Fixed in: 0.8.4
vLLM allows Remote Code Execution by Pickle Deserialization via AsyncEngineRPCServer() RPC server entrypoints
vLLM deserialization vulnerability in vllm.distributed.GroupCoordinator.recv_object
vLLM Deserialization of Untrusted Data vulnerability
vLLM Allows Remote Code Execution via Mooncake Integration
Fixed in: 0.8.0, 288ca110f68d23909728627d3100e5a8db820aa2
vLLM denial of service via outlines unbounded cache on disk
Fixed in: 0.8.0
vLLM uses Python 3.12 built-in hash() which leads to predictable hash collisions in prefix cache
Fixed in: 0.7.2, 432117cd1f59c76d97da2eaff55a7d758301dbc7
vllm: Malicious model to RCE by torch.load in hf_model_weights_iterator
Fixed in: 0.7.0, d3d6bb13fb62da3234addf6574922a4ec0513d04
vLLM denial of service vulnerability
Fixed in: 0.5.5
vLLM Denial of Service via the best_of parameter