[ctx]ctxseccontext security: attacks on the harness and context layer of AI agents, plus exploited-now news, hunting and playbooks.

← Threat Hunting · markdown

official (SANS Internet Storm Center diary, Nov 16, 2026; LiteLLM advisory) · published here

SANS honeypots see LiteLLM/Starlette RCE chain probes with malformed Host headers like 'localhost]' on /mcp-rest/test/tools/list

  • Hunt: POSTs to /mcp-rest/test/tools/list; Host headers containing [, ], /, ? or #, or values such as localhost], [localhost, 127.0.0.1], litellm.proxy; request bodies with "transport": "stdio" and a command; user agent Python/3.14 aiohttp/3.14.5.
  • Fix: LiteLLM 1.84.0 or later (or the backported patch releases) and Starlette 1.0.1 or later.
  • WAF lesson from SANS: block characters that can't appear in a hostname, or allow only expected hostnames, instead of matching the PoC string.

Sources: SANS ISC diary 33522: Scans for LiteLLM/Starlette (CVE-2026-42271, CVE-2026-48710) (Nov 16, 2026) · LiteLLM: Authentication bypass via Host header injection (GHSA-4xpc-pv4p-pm3w), fixed in 1.84.0+