Vulnerability database · updated every 15 minutes
New vulnerabilities in the software websites run.
CVEs and security advisories for WordPress, its plugins and themes, npm, PyPI and the servers behind websites, merged from NVD, the CVE Program, GitHub, OSV.dev and WPVulnerability, with CISA's list of the ones attackers exploit.
- Tracked
- 55,370
- Published in 7 days
- 1,871
- Critical in 30 days
- 488
- Actively exploited
- 542
Results for “vLLM”
169 vulnerabilities- High
GHSA-mm38-5g96-7j6x · vLLM
Nemotron-VL processors disable Pillow decompression-bomb guard globally, enabling unauthenticated DoS
Published 9 Oct 2026 · Fixed in 0.25.0 · Project advisories (GitHub) - High
GHSA-vfp2-c8pq-v6h6 · vLLM
SSRF via X-Request-Id header drives P2pNcclConnector outbound ZMQ+NCCL pair, leaking KV-cache tensors
Published 9 Oct 2026 · Fixed in 0.24.0 · Project advisories (GitHub) - Medium 6.5
GHSA-r45p-7qwx-qwjw · vLLM
Multimodal chat audio decoding bypasses the decompression-bomb duration limit
Published 9 Oct 2026 · Fixed in 0.24.0 · Project advisories (GitHub) - Medium 5.9
GHSA-w2v2-5662-8v5q · vLLM
Rust gRPC Generate accepts out-of-vocabulary prompt token IDs and can terminate EngineCore
Published 9 Oct 2026 · Fixed in 0.24.0 · Project advisories (GitHub) - Medium 5.9
GHSA-6pgw-f3pv-4h8q · vLLM
Rust gRPC output_candidates.token_ids bypasses logprob_token_ids limits and can terminate EngineCore
Published 9 Oct 2026 · Fixed in 0.24.0 · Project advisories (GitHub) - Medium 5.9
GHSA-rwfw-xvg3-3937 · vLLM
Rust gRPC CandidateTokens.top_n bypasses max_logprobs and can terminate EngineCore
Published 9 Oct 2026 · Fixed in 0.24.0 · Project advisories (GitHub) - High 7.5
GHSA-8c65-hq7q-r7jm · vLLM
Remote DoS in vLLM via allowed_token_ids Speculative-Decoding Metadata Mismatch
Published 9 Oct 2026 · Fixed in 0.18.0 · Project advisories (GitHub) - High 7.7
GHSA-rhcx-5729-88vg · vLLM
Cross-tenant DoS via uninitialized LogprobsTensors with prompt_logprobs
Published 9 Oct 2026 · Fixed in 0.31.0 · Project advisories (GitHub) - Low 3.7
GHSA-6cxc-2vcg-w5qc · vLLM
V1 InputBatch.condense leaves a stale allowed_token_ids mask on recycled batch rows
Published 6 Oct 2026 · Fixed in 0.31.0 · Project advisories (GitHub) - Low 2.1
CVE-2026-105922 · vLLM
vllm-project vLLM Penalty utils.py get_token_bin_counts_and_mask denial of service
Published 6 Oct 2026 · NVD, CVE Program - High 7.1
CVE-2026-114050 · vLLM
Forged multimodal UUID + P0/P1 processor-cache eviction drift: unauthenticated remote crash of the whole vLLM engine
Published 6 Oct 2026 · Fixed in 0.31.0 · NVD, CVE Program, Project advisories (GitHub) - Medium 6.5
GHSA-rh6f-3x46-j33q · vLLM
Caller-controllable values become or collide with shared cache keys, enabling cross-request cache aliasing (2 sites: multimodal EXIF hash and prefix-cache extra keys)
Published 6 Oct 2026 · Fixed in 0.31.0 · Project advisories (GitHub) - High 8.1
GHSA-h3rc-6mm3-gc2m · vLLM
Request-controlled mm_processor_kwargs.code_revision allows remote code execution
Published 6 Oct 2026 · Fixed in 0.31.0 · Project advisories (GitHub) - High 7.5
GHSA-gx7p-2j49-hfq4 · vLLM
DoS via unvalidated mm_processor_kwargs in multimodal image processors (Isaac, InternVL, H2OVL, etc.)
Published 6 Oct 2026 · Fixed in 0.31.0 · Project advisories (GitHub) - High 8.7
CVE-2026-114052 · vLLM
MOSS-Audio request-controlled processor cache permits remote memory exhaustion
Published 6 Oct 2026 · Fixed in 0.31.0 · NVD, CVE Program, Project advisories (GitHub) - Medium 6.5
GHSA-4xqp-c3mv-qff7 · vLLM
An oversized min_tokens in one request hangs the vLLM engine while /health stays green
Published 6 Oct 2026 · Fixed in 0.31.0 · Project advisories (GitHub) - Medium 5.3
CVE-2026-105775 · vLLM
vllm-project vLLM Completions Request mamba_mixer2.py conv_ssm_forward out-of-bounds
Published 6 Oct 2026 · CVE Program - Medium 5.5
CVE-2026-103241 · vLLM
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…
Published 30 Sept 2026 · NVD - Medium 6.5
CVE-2026-105753 · 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
Published 28 Sept 2026 · Fixed in 0.28.0 · NVD, CVE Program, Project advisories (GitHub) - Medium 6.5
GHSA-m52c-39rh-f3gp · vLLM
Crafted IAMF audio upload reaches a PyAV/FFmpeg native heap overflow through the speech transcription path — denial of service
Published 28 Sept 2026 · Fixed in 0.29.0 · Project advisories (GitHub) - Low 3.1
CVE-2026-105752 · vLLM
Harmony tool continuations drop `cache_salt` — restoring a cross-tenant prefix-cache membership oracle
Published 28 Sept 2026 · Fixed in 0.30.0 · NVD, CVE Program, Project advisories (GitHub) - High 7.1
CVE-2026-100654 · vLLM
vLLM before 0.29.0 accepts user-controlled stop_token_ids on the OpenAI-compatible POST /v1/completions and POST /v1/chat/completions endpoints but validates only that the values are integers, not that each token id is within the model vocabulary/logits range. When min_tokens > 0, the stop token ids are used as logits indices to suppress stop tokens, so an out-of-range id reaches a CUDA indexing…
Published 26 Sept 2026 · Fixed in 0.29.0 · NVD - High 8.3
CVE-2026-100653 · vLLM
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…
Published 26 Sept 2026 · Fixed in 0.28.0 · NVD - High 8.2
CVE-2026-100652 · vLLM
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…
Published 26 Sept 2026 · Fixed in 0.24.0 · NVD - High 7.1
CVE-2026-100651 · vLLM
vLLM before 0.29.0 fails to enforce decoder prompt-length validation on the disaggregated serving endpoint /inference/v1/generate. When the request contains a 'features' (multimodal) payload, vllm/entrypoints/serve/disagg/serving.py builds a multimodal EngineInput directly from the caller-supplied token_ids, and GenerateRequest.token_ids (vllm/entrypoints/serve/disagg/protocol.py) is not checked…
Published 26 Sept 2026 · Fixed in 0.29.0 · NVD - High 7.1
CVE-2026-100650 · vLLM
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…
Published 26 Sept 2026 · NVD - Medium 5.3
CVE-2026-105758 · vLLM
Qwen2-VL / Qwen3-VL video samplers bound on request-controlled max_frames, which the num_frames ceiling does not reach
Published 23 Sept 2026 · Fixed in 0.30.0 · NVD, CVE Program, Project advisories (GitHub) - High 7.5
CVE-2026-105759 · vLLM
Unbounded Prometheus label cardinality from attacker-controlled HTTP method tokens in the vLLM Rust frontend metrics middleware (unauthenticated denial of service)
Published 23 Sept 2026 · Fixed in 0.30.0 · NVD, CVE Program, Project advisories (GitHub) - Medium 5.3
CVE-2026-105760 · vLLM
GLMGA video sampling permits request-driven CPU and memory exhaustion
Published 23 Sept 2026 · Fixed in 0.30.0 · NVD, CVE Program, Project advisories (GitHub) - Medium 6.5
CVE-2026-105754 · vLLM
Scale-out disaggregated multimodal transport trusts caller-supplied features — shared EngineCore denial of service, encoder-cache poisoning, and transport integrity loss (5 sites)
Published 23 Sept 2026 · Fixed in 0.30.0 · NVD, CVE Program, Project advisories (GitHub) - Medium 6.5
CVE-2026-105757 · vLLM
Structured-output request errors escape the request boundary and terminate the shared EngineCore — engine-fatal denial of service (3 sites)
Published 23 Sept 2026 · Fixed in 0.30.0 · NVD, CVE Program, Project advisories (GitHub) - Medium 4.2
CVE-2026-105755 · 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`
Published 23 Sept 2026 · Fixed in 0.30.0 · NVD, CVE Program, Project advisories (GitHub) - Medium 6.5
CVE-2026-105756 · vLLM
Loose `cache_salt` validation lets a single request kill EngineCore on LMCache-MP deployments — uncaught downstream `ValueError` denial of service
Published 23 Sept 2026 · Fixed in 0.30.0 · NVD, CVE Program, Project advisories (GitHub) - High 8.7
CVE-2026-94627 · vLLM
vLLM Mooncake connector through 0.29.0 fails to properly manage GPU KV cache block ownership when concurrent child requests share a single transfer ID in prefill/decode disaggregated deployments. Attackers can trigger GPU memory exhaustion by submitting completion requests with multiple prompts, causing orphaned KV cache blocks to accumulate until process restart and eventually preventing…
Published 21 Sept 2026 · NVD - High 8.7
CVE-2026-94626 · vLLM
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.
Published 21 Sept 2026 · NVD - Medium 6.9
CVE-2026-94625 · vLLM
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.
Published 21 Sept 2026 · NVD - High 8.7
CVE-2026-94624 · vLLM
vllm p2p never connected zmq_exhaustion
Published 21 Sept 2026 · NVD, Project advisories (GitHub) - High 8.7
CVE-2026-94623 · vLLM
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…
Published 21 Sept 2026 · NVD - High 8.7
CVE-2026-94622 · vLLM
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.
Published 21 Sept 2026 · NVD - Low 2.3
CVE-2026-93989 · vLLM
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.
Published 19 Sept 2026 · NVD - Medium 6.3
CVE-2026-93841 · vLLM
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…
Published 18 Sept 2026 · NVD - Medium 6.3
CVE-2026-93840 · vLLM
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.
Published 18 Sept 2026 · Fixed in 0.29.0 · NVD - High 8.7
CVE-2026-93436 · vLLM
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.
Published 17 Sept 2026 · NVD - Medium 6.5
GHSA-hhv2-872h-628q · vLLM
Incomplete artifact pin propagation in FunAudioChat and Tarsier2
Published 12 Sept 2026 · Fixed in 0.28.0 · Project advisories (GitHub) - Medium 6.5
GHSA-v5gm-qgmv-gc6c · vLLM
Out-of-range `stop_token_ids` with `min_tokens` can kill vLLM EngineCore
Published 12 Sept 2026 · Fixed in 0.29.0 · Project advisories (GitHub) - Medium 5.9
GHSA-qff2-492f-9fm4 · vLLM
Rust HTTP/gRPC stop_token_ids bypass Python vocab-bound fix and can terminate EngineCore
Published 12 Sept 2026 · Fixed in 0.24.0 · Project advisories (GitHub) - Medium 6.5
GHSA-3mqx-f33v-vgp9 · vLLM
Disaggregated generate skips decoder prompt-length validation for some multimodal processors
Published 12 Sept 2026 · Fixed in 0.29.0 · Project advisories (GitHub) - Medium 6.3
CVE-2026-100649 · vLLM
Sampler Subclass Counter Shadowing Bypasses PyNvVideoCodec Decoder Limits and GPU Memory Accounting
Published 12 Sept 2026 · Fixed in 0.29.0 · NVD, Project advisories (GitHub) - Medium 6.5
GHSA-p6g9-7v3x-m8mv · vLLM
Remote media is fetched and fully materialized before the documented media size / item limits are enforced — remote pre-inference memory and bandwidth exhaustion (4 sites)
Published 12 Sept 2026 · Fixed in 0.29.0 · Project advisories (GitHub) - Medium 6.9
CVE-2026-100648 · vLLM
Uncontrolled resource consumption: multimodal chat audio decoding ignores `VLLM_MAX_AUDIO_CLIP_FILESIZE_MB`
Published 12 Sept 2026 · Fixed in 0.29.0 · NVD, Project advisories (GitHub)
Where it comes from
The list holds every advisory UpHeron reads for the software it watches: WordPress and every plugin and theme its plugin reports, the packages in monitored repositories and manifests, and the servers and frameworks found on monitored websites. One vulnerability from several sources is one entry, under its CVE.
- NVD, the US National Vulnerability Database. This product uses the NVD API but is not endorsed or certified by the NVD.
- New and changed records of the CVE Program, read every 15 minutes, so a rating or fix added after a CVE is published reaches the list, and its alerts, within the hour. CVE content © 1999–2026 The MITRE Corporation, used under the CVE terms of use; CVE® is a registered trademark of The MITRE Corporation.
- OSV.dev, with the GitHub Advisory Database (CC BY 4.0).
- WPVulnerability (CC0) for WordPress plugins and themes, and WordPress's own security advisories.
- Node.js's own security release posts, read every 15 minutes, which name its vulnerabilities the day they are fixed, days before their CVE records.
- The security advisories projects publish on GitHub, and the CVEs libraries such as Brotli name in their release notes, which also count for the runtimes that bundle them.
- CISA's Known Exploited Vulnerabilities catalog.
Each day's new ones, grouped, and every critical one as soon as we find it: on the blog.
Run WordPress? The UpHeron plugin alerts you the day one of these affects your own site.