VULNERABILITY OVERVIEW
A command injection flaw (CWE-77/78) in LiteLLM's MCP server preview endpoints (POST /mcp-rest/test/connection and POST /mcp-rest/test/tools/list) allows any authenticated user — including holders of low-privilege internal-user API keys — to spawn arbitrary OS commands as subprocesses on the proxy host. Horizon3.ai discovered that chaining CVE-2026-42271 with CVE-2026-48710 (a Starlette BadHost header validation bypass) removes the authentication requirement entirely, yielding a combined CVSS 10.0 unauthenticated RCE chain. Successful exploitation exposes all LLM provider API keys (OpenAI, Anthropic, Google, Azure, AWS Bedrock) and AI infrastructure connected through the proxy. CISA added to KEV on June 8, 2026, with a June 22 BOD 22-01 federal deadline. Public PoC and detailed chain analysis are available on GitHub.
CVSS BREAKDOWN
LiteLLM versions 1.74.2 through 1.83.6 (fixed in 1.83.7-stable)CITATIONS
- → https://thehackernews.com/2026/06/litellm-flaw-cve-2026-42271-exploited.html
- → https://github.com/BerriAI/litellm/security/advisories/GHSA-v4p8-mg3p-g94g
- → https://www.cisa.gov/known-exploited-vulnerabilities-catalog
- → https://nvd.nist.gov/vuln/detail/CVE-2026-42271