{
 "id": "AIR-2026-0848",
 "slug": "cve-2026-59821-litellm-custom-code-guardrails-bypass-sandbox",
 "title": "CVE-2026-59821: LiteLLM Custom Code Guardrails Bypass Sandboxing",
 "summary": "LiteLLM custom code guardrail creation paths failed to apply proper sandboxing and validation mechanisms. This flaw allowed privileged users to execute arbitrary Python code in the proxy container.",
 "type": "attack_on_ai",
 "categories": [
  "infra_vuln"
 ],
 "severity": 2,
 "targets": [
  "framework_or_tooling"
 ],
 "products": [
  "LiteLLM"
 ],
 "incident_date": "2026-07-22",
 "incident_date_precision": "day",
 "first_reported": "2026-07-08",
 "sources": [
  {
   "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-59821",
   "publisher": "NVD",
   "title": "CVE-2026-59821: LiteLLM is a proxy server (AI Gateway) to call LLM APIs in OpenAI (or native) format.",
   "published_at": "2026-07-08"
  },
  {
   "url": "https://github.com/advisories/GHSA-72m8-9m7m-h278",
   "publisher": "GitHub Advisory Database",
   "title": "CVE-2026-59821: LiteLLM: Custom Code Guardrails production endpoints bypass code safety checks",
   "published_at": "2026-07-22"
  }
 ],
 "source_count": 2,
 "confidence": 0.95,
 "status": "published",
 "origin": "seed_ghsa",
 "tags": [
  "CVE-2026-59821",
  "litellm",
  "rce",
  "guardrail_bypass"
 ],
 "created_at": "2026-10-18T07:00:14.101Z",
 "updated_at": "2026-10-18T10:04:21.885Z"
}