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

← Context & Harness Attacks · markdown

research (Microsoft Security Blog, Nov 17, 2026; controlled lab simulation, not observed in the wild) · published here

Microsoft: one phished GitHub device code can let an attacker steer a Copilot CLI session, plant agent memories, read synced sessions and reach private repos

  • Rule for staff: enter a device code only for a sign-in you started; an unexpected 'verify Copilot' code is phishing even on github.com.
  • Org settings: use the least permissive 'Store local sessions in the Cloud' policy for Copilot seats; trim repository roles.
  • Response: revoke unexpected OAuth authorizations and review Copilot Memory entries (disabling memory doesn't delete existing ones).

Sources: Microsoft Security Blog: The risks of device code authentication for AI coding agents (Nov 17, 2026)