HighVulnerabilityLLM-specific

CVE-2026-22807: vLLM arbitrary code execution through Hugging Face auto_map model loading

Source: NVD(opens in a new tab)
Identifier: CVE-2026-22807
Published: January 21, 2026
Record updated: October 9, 2026

Affected: vllm >= 0.10.1, < 0.14.0
Fixed in: 0.14.0
released January 20, 2026, 2 days before disclosure
Known exploitation: Not listed in the CISA Known Exploited Vulnerabilities catalog at the last check.
EPSS: 1.0%

Summary

vLLM versions from 0.10.1 up to but not including 0.14.0 load Hugging Face `auto_map` dynamic modules during model resolution without checking `trust_remote_code`. An attacker who can influence the model repo or path, whether a local directory or a remote Hugging Face repo, can run arbitrary Python code on the vLLM host at server startup, before any request handling and without API access.

Mitigation

Fixed in 0.14.0.

Details

Scores and exploitation

CVSS score: 8.8(high)
Patch available: Yes
EPSS: 1.0%Estimated probability that this CVE is exploited in the next 30 days, from the FIRST Exploit Prediction Scoring System. Read on November 4, 2026.

Weaknesses and techniques

Weakness (CWE): CWE-94
Attack pattern (CAPEC): CAPEC-242
AI threat technique (MITRE ATLAS): AML.T0010

Affected

Packages: vllm@>= 0.10.1, < 0.14.0 (fixed: 0.14.0)

Classification

Attack type: Supply chain
Attack sophistication: Moderate
AI component targeted: Inference
Impact: Confidentiality,
Integrity,
Availability
Labels: security

How this record was produced

Summaries, mitigations and labels on this page are generated automatically from the source. The linked source is authoritative. The source gives this advisory a description and no title, so the title above is a headline generated from the source text.

Original source: https://nvd.nist.gov/vuln/detail/CVE-2026-22807

First tracked: February 15, 2026 at 08:44 PM ET, from NVD/CVE Database

Classifier: claude-haiku-5-5, prompt v4

Classifier confidence: 95% (self-assessed)

Affected packages in the Exposure Registry

Matched by package name and ecosystem. Each entry shows whether the package delegates to a language model and how many tracked packages depend on it.

vllmPyPILLM dependency since 2023-08-25 · 15 tracked dependents

Topics

Inference infrastructure

Summary

Mitigation

Affected packages in the Exposure Registry

Topics

Related items