vLLM Vulnerabilities

pipLLM Inference

AI Threat Alert tracks 188 known vulnerabilities in vLLM, 13 rated critical — an AI/ML llm inference in the pip ecosystem. Each CVE includes CVSS severity, EPSS exploit probability, patch status, and CISO-grade analysis.

61
Risk Score
188
Total CVEs
13
Critical
pip
Ecosystem
Nov 18, 2026
Last CVE
26%
Patch Rate
37d
Avg Time to Patch
96,942 stars24,834 forks10,395 issues93 dependentsLast push Nov 15, 2026
View on GitHub

Known Vulnerabilities(188 total, page 1 of 8)

SeverityCVE IDSummaryCVSSPublished
HIGHCVE-2026-130686vLLM from 0.6.0 before 0.33.0 lacks a request-wide media budget, allowing remote inference clients to exhaust frontend resources by submitting many remote image URLs in one request. Attackers can send a compact request to /v1/messages or /v1/responses that forces downloading and decoding of every image before prompt-length admission, exhausting memory, CPU, and connections.7.5Nov 18, 2026HIGHCVE-2026-130685vLLM from 0.28.0 before 0.33.0 contains an improper input validation vulnerability that allows unauthenticated attackers to crash the shared EngineCore by sending out-of-range prompt token ids when served with --skip-tokenizer-init. Attackers can submit a large positive token id to /v1/completions, /v1/embeddings, or /pooling, bypassing the tokenizer-gated vocab check and causing a fatal embedding index error that denies service to all tenants.7.5Nov 18, 2026MEDIUMCVE-2026-130684vLLM before 0.33.0 contains a path traversal vulnerability that allows malicious model publishers to redirect the served model and tokenizer by setting speculators_config verifier.name_or_path in config.json. Attackers can point this field at another local checkpoint directory to leak its tokenizer vocabulary via unauthenticated /detokenize requests and gain query access to private models.5.4Nov 18, 2026MEDIUMCVE-2026-129356vLLM: race condition in KV cache prefix handler5.0Nov 15, 2026MEDIUMCVE-2026-126399vLLM: num_frames bypass causes memory exhaustion DoS6.5Nov 13, 2026MEDIUMCVE-2026-126398vLLM: reachable assertion in XD-RoPE crashes engine5.9Nov 13, 2026HIGHCVE-2026-126397vLLM: unbounded max_patches enables CPU exhaustion DoS7.5Nov 13, 2026MEDIUMCVE-2026-126396vLLM: unbounded min_pixels in MiniMax M3 VL causes DoS5.3Nov 13, 2026MEDIUMCVE-2026-126395vLLM: heap address leak via error message redaction5.3Nov 13, 2026MEDIUMCVE-2026-126394vLLM: unbounded num_frames in openpangu video OOM-kills API5.3Nov 13, 2026MEDIUMCVE-2025-13327uv: ZIP parsing differential allows code execution6.3Feb 27, 2026MEDIUMCVE-2026-115918vLLM: chat template bypass enables DoS via /v1/responses6.5Oct 26, 2026LOWCVE-2026-115917vLLM: wildcard CORS lets web pages reach local servers3.4Oct 26, 2026MEDIUMCVE-2026-115914vLLM: zero-sum OCR image crashes engine (DoS)6.5Oct 26, 2026MEDIUMCVE-2026-115916vLLM: off-by-one token ID check crashes engine (DoS)6.5Oct 26, 2026HIGHCVE-2026-115915vLLM: image decoder abuse enables RCE/DoS8.1Oct 26, 2026MEDIUMCVE-2026-115913vLLM: Phi-4 audio attention causes GPU OOM DoS6.5Oct 26, 2026HIGHCVE-2026-115912vLLM: pickle deserialization RCE via multimodal cache7.5Oct 26, 2026MEDIUMCVE-2026-115911vLLM: chat template DoS via unbounded Jinja rendering6.5Oct 26, 2026MEDIUMCVE-2026-116181vLLM: invalid stop_token_ids crash engine, GPU OOB write6.5Oct 25, 2026HIGHCVE-2026-116180vLLM: uninitialized tensor lets tenant crash engine7.7Oct 25, 2026HIGHCVE-2026-116179vLLM: num_frames override enables memory-exhaustion DoS7.5Oct 25, 2026HIGHCVE-2026-116178vLLM: multimodal metadata DoS crashes inference engine7.5Oct 25, 2026HIGHCVE-2026-116177vLLM: decompression bomb in Nemotron-VL causes DoS7.5Oct 25, 2026HIGHCVE-2026-116176vLLM: SSRF via X-Request-Id leaks NCCL KV-cache data8.2Oct 25, 2026

Showing 1–25 of 188

Frequently asked questions

What is vLLM?

vLLM is an AI/ML llm inference tracked by AI Threat Alert for security vulnerabilities in the pip ecosystem.

How many known vulnerabilities does vLLM have?

vLLM has 188 known CVEs, 13 of them critical, tracked from NVD and GitHub Advisory.

Which ecosystem is vLLM distributed in?

vLLM is distributed via the pip ecosystem and categorized as llm inference.

Where does the vLLM vulnerability data come from?

Vulnerability data is sourced from NVD and GitHub Advisory, enriched with CVSS, EPSS, exploit signals, and patch status for each CVE.

How do I assess the risk of vLLM?

Review each CVE below — every entry shows CVSS severity, EPSS exploit probability, exploitation signals, and whether a patched version is available.

Monitor vLLM in your stack

Get instant alerts when new vulnerabilities affect vLLM. CISO analysis, ATLAS technique mappings, and compliance reports included.

Start Monitoring