CISA Adds Seven Vulnerabilities to Catalog, Three Targeting AI Infrastructure

CISA added seven vulnerabilities to its Known Exploited Vulnerabilities catalog on September 2, 2026. Three of these vulnerabilities target AI and machine learning infrastructure, making up nearly half of the additions.

The three AI-specific CVEs attack distinct layers of the stack. CVE-2026-59822 affects LiteLLM, an AI gateway and proxy, allowing unauthenticated Model Context Protocol sessions via arbitrary Bearer tokens. This impacts all versions prior to 1.84.0 with a CVSS score of 8.8.

CVE-2026-48710 targets Starlette/FastAPI, the ASGI framework underlying vLLM, LiteLLM, and MCP servers. The 'BadHost' vulnerability allows Host header injection, bypassing path-based authentication middleware with malformed characters in the Host header.

The remaining four CVEs address persistent enterprise threats. CVE-2026-49869 in Kestra OSS carries a CVSS 10.0 for a suffix-match authentication bypass allowing any path ending in '/configs' to skip authentication entirely, yielding unauthenticated RCE as root. This requires remediation by September 5.