[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 Cursor's Nov 13 incident update and Dokploy's Nov 13 notice

Tip: any key you ever pasted into a coding agent, AI chat or support ticket may be sitting in someone's logs. Rotate it and stop pasting

  • Rotate: API keys, tokens and passwords that were ever pasted into an agent session, AI chat, support ticket or hosted panel. Revoke at the issuing provider; deleting the message doesn't invalidate the key.
  • Check usage: review the provider's usage and audit logs for calls you don't recognize, especially LLM API spend (stolen AI keys are resold).
  • Stop the habit: give agents scoped, short-lived credentials via environment or a secrets manager, not chat; use privacy or no-training modes where offered; add secret scanning to anything that stores transcripts.
  • Vendor review: ask AI vendors whether session data is shared with third parties for training or inference, and whether it is scrubbed of secrets first.

Sources: Cursor: An update on our recent security incident (Nov 13, 2026) · Dokploy: Security notice: AI provider API keys on Dokploy Cloud (Nov 13, 2026)