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

← AI · markdown

analysis and opinion (VulnCheck blog by Patrick Garrity, Nov 9, 2026); figures from his tracking repo and Anthropic's ledger as of Nov 2

VulnCheck's seven-month audit of Anthropic's Glasswing: 442 CVEs, 2 known exploited, 217 ledger entries marked fixed

  • Caveat: the CVE count and the 'fixed' count are different measures. Anthropic's own dashboard (Nov 2) says 629 of 5,087 disclosed vulnerabilities have been patched to its knowledge and 230 CVE records issued, so the numbers depend on which status field you use.
  • For defenders: treat AI-found CVEs like any others and prioritize by exploitation, not by who found them; for maintainers, ask AI-scanning vendors for reproducible PoCs before triage.

Sources: VulnCheck: After 7-months, Glasswing Overwhelms Maintainers with Underwhelming Findings (Nov 9, 2026) · LWN: Glasswing overwhelms maintainers with underwhelming findings