Log in Start free

Vulnerability · Published Mar 26, 2026 · Assigned by GitHub_M

CVE-2026-27893vLLM's hardcoded trust_remote_code=True in NemotronVL and KimiK25 bypasses user security opt-out

High 8.8

In plain English

A high-severity security flaw in vllm from vllm-project. There are no confirmed reports of exploitation so far. The chance it is exploited in the next 30 days is estimated at 1.4% (higher than 72% of all known vulnerabilities). Fixed versions are available - upgrade the affected package.

Severity (CVSS 3.1)8.8High
Exploit odds (30 days)1.4%top 28.0% of all CVEs
Exploited in the wildNo reportsCISA SSVC: none
Attack toolingNoneno public tooling known

Official description

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 user's explicit `--trust-remote-code=False` security opt-out. This enables remote code execution via malicious model repositories even when the user has explicitly disabled remote code trust. Version 0.18.0 patches the issue.

Affected packages & fixed versions

EcosystemPackageFixed inAdvisory
PyPIvllm0.18.0GHSA-7972-pg2x-xr59
PyPIvllm0.18.0PYSEC-2026-2297

Affected products (as reported)

VendorProductVersions
vllm-project vllm >= 0.10.1, < 0.18.0

Weakness: CWE-693, CWE-501 · Vector: CVSS:3.1/AV:N/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H

References