What happened
AWS Security Blog (Nov 13, 2026) published Amazon Threat Intelligence analysis of threat actors running cybercrime campaigns with ARTEX, an open-source agentic pentest framework. From Oct 3 to Nov 5, 2026, Amazon's MadPot decoy-sensor network captured AI-agent requests whose operating memory revealed stolen credentials and deepfake-fraud attempts from prior offensive campaigns, and AWS notified affected customers. The FBI also issued a flash alert on ARTEX weaponization.
Why it matters
Primary-vendor evidence that AI agents are now being used as autonomous attack tools — and that their persistent operating memory can leak campaign details to defenders, informing agent-memory hygiene and agentic-attack detection for security teams.
Applicability
Detection and threat-intel teams should monitor for ARTEX/agentic pentest tooling and apply the agent-memory/credential-leakage observations to guidance on securing their own AI agents and hunting AI-agent-driven intrusions.