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

← Tips & Guides · markdown

guidance based on IBM's ContextForge CVE-2026-110279 bulletin

Tip: if your MCP or API gateway injects upstream credentials, strip every client-supplied auth header before forwarding

  • Default deny: at the edge, remove Authorization, Proxy-Authorization, X-Upstream-* and any custom auth headers from client requests unless a route explicitly needs them.
  • Set, don't merge: have the gateway set the upstream credential last, overwriting anything already present.
  • Test it: send a request with a bogus upstream auth header and confirm upstream logs show the gateway's identity, not yours.

Sources: IBM security bulletin: ContextForge MCP Gateway CVE-2026-110279 (Nov 12, 2026)