Five Eyes spook shops warn rapid rollouts of agentic AI are too risky

Five Eyes national security and cybersecurity agencies jointly published guidance cautioning that agentic AI systems introduce expanded attack surfaces and can behave unexpectedly; they recommend slow, security-first adoption, strong governance, human oversight, and prioritizing resilience and reversibility. The guidance illustrates misuse scenarios (e.g., over-privileged agents enabling fraud), enumerates 23 risk categories and 100+ best practices, and urges vendors, developers and defenders to improve testing, monitoring and threat intelligence for agentic AI.