CVE-2026-27893: vLLM Hardcoded Trust Override Enables RCE

# CVE-2026-27893: vLLM Hardcoded Trust Override Enables RCE

https://gcdradar.vercel.app/incidents/cve-2026-27893-vllm-hardcoded-trust-override-enables-rce · AIR-2026-0529 · closed 2026-03-27

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Fix status unknown
- First reported: 2026-03-27
- AI role: Attack on AI
- Products: vLLM
- CVE ids: CVE-2026-27893
- CVSS: 8.8

## Summary

Specific model implementation files in vLLM hardcoded trust parameters that bypassed user opt-out configurations, permitting remote code execution via untrusted model repositories. Patches were developed to respect security constraints.

## What to do

No action recorded; read the sources below and the vendor's advisory.

## Latest updates

- 2026-03-27: CVE-2026-27893: vLLM is an inference and serving engine for large language models (LLMs). (https://nvd.nist.gov/vuln/detail/CVE-2026-27893)
- 2026-03-27: CVE-2026-27893: vLLM has Hardcoded Trust Override in Model Files Enables RCE Despite Explicit User Opt-Out (https://github.com/advisories/GHSA-7972-pg2x-xr59)

## Sources

- 2026-03-27 GitHub Advisory Database: CVE-2026-27893: vLLM has Hardcoded Trust Override in Model Files Enables RCE Despite Explicit User Opt-Out https://github.com/advisories/GHSA-7972-pg2x-xr59
- 2026-03-27 NVD: CVE-2026-27893: vLLM is an inference and serving engine for large language models (LLMs). https://nvd.nist.gov/vuln/detail/CVE-2026-27893

Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.