{
 "id": "AIR-2026-0600",
 "slug": "cve-2026-42271-litellm-authenticated-command-execution-via-m",
 "title": "CVE-2026-42271: LiteLLM Authenticated Command Execution via MCP Endpoints",
 "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.",
 "type": "attack_on_ai",
 "categories": [
  "infra_vuln"
 ],
 "severity": 2,
 "targets": [
  "framework_or_tooling"
 ],
 "products": [
  "LiteLLM"
 ],
 "incident_date": "2026-04-25",
 "incident_date_precision": "day",
 "first_reported": "2026-04-25",
 "sources": [
  {
   "url": "https://github.com/advisories/GHSA-v4p8-mg3p-g94g",
   "publisher": "GitHub Advisory Database",
   "title": "CVE-2026-42271: LiteLLM: Authenticated command execution via MCP stdio test endpoints",
   "published_at": "2026-04-25"
  },
  {
   "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-42271",
   "publisher": "NVD",
   "title": "CVE-2026-42271: LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format.",
   "published_at": "2026-05-08"
  }
 ],
 "source_count": 2,
 "confidence": 0.95,
 "status": "published",
 "origin": "seed_ghsa",
 "tags": [
  "CVE-2026-42271",
  "command_injection",
  "mcp"
 ],
 "created_at": "2026-10-18T07:00:14.101Z",
 "updated_at": "2026-10-18T10:04:21.885Z"
}