CVE-2026-42271: LiteLLM Authenticated Command Execution via MCP Endpoints

# CVE-2026-42271: LiteLLM Authenticated Command Execution via MCP Endpoints

https://gcdradar.vercel.app/incidents/cve-2026-42271-litellm-authenticated-command-execution-via-m · AIR-2026-0600 · closed 2026-04-25

- Lane: Patch now
- Severity: 2 (Disclosed vulnerability)
- Fix status: Fix status unknown
- First reported: 2026-04-25
- AI role: Attack on AI
- Products: LiteLLM
- CVE ids: CVE-2026-42271
- CVSS: 8.8

## Summary

LiteLLM preview endpoints for Model Context Protocol servers accepted arbitrary command parameters in request bodies without adequate validation. Authenticated users could spawn arbitrary subprocesses on the proxy host.

## What to do

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

## Latest updates

- 2026-05-08: CVE-2026-42271: LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. (https://nvd.nist.gov/vuln/detail/CVE-2026-42271)
- 2026-04-25: CVE-2026-42271: LiteLLM: Authenticated command execution via MCP stdio test endpoints (https://github.com/advisories/GHSA-v4p8-mg3p-g94g)

## Sources

- 2026-04-25 GitHub Advisory Database: CVE-2026-42271: LiteLLM: Authenticated command execution via MCP stdio test endpoints https://github.com/advisories/GHSA-v4p8-mg3p-g94g
- 2026-05-08 NVD: CVE-2026-42271: LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format. https://nvd.nist.gov/vuln/detail/CVE-2026-42271

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