Government Alliance Confronts Reality of Unpredictable Agentic AI Already Deployed in Critical Systems

The Five Eyes security alliance (CISA, NSA, and counterparts) released the first coordinated policy specifically targeting agentic AI, framing autonomous systems not as a future risk but as already behaving unpredictably within live government and infrastructure deployments. The most significant admission: existing threat assessment frameworks (MITRE ATLAS, OWASP) lag dangerously behind multi-step autonomous agents operating with minimal oversight, revealing a material gap between known threats and available evaluation tools. This candid confession signals practitioners need zero-trust deployment practices immediately rather than waiting for security standards to mature.