CISA Warning: LiteLLM Flaw Could Expose Enterprise AI Gateways

On June 9 CISA added CVE-2026-42271 — a command-injection flaw in LiteLLM's MCP endpoints — to its Known Exploited Vulnerabilities catalog, requiring rapid patching; exploitation can lead to remote code execution and, when chained with CVE-2026-48710 (an authentication bypass), to unauthenticated RCE. The report notes active weaponization, emphasizes that AI gateways are high-value attack surfaces with broad downstream access, and recommends applying service-account governance (least privilege, short-lived credentials, platform-layer enforcement, and immutable audit logs) to mitigate systemic risk.