Beyond the Chatbot: Why NIST is Rewriting the Rules for Autonomous AI

The article analyzes NIST’s RFI on AI Agent Systems, arguing that the risk landscape has shifted from conversational AI to autonomous agents capable of affecting external state, and highlighting novel threats such as indirect prompt injection, data poisoning/backdoors, and specification gaming. It emphasizes national security stakes (critical infrastructure and potential CBRNE implications), calls for adapting zero-trust principles (least privilege, instruction hierarchies, rollbacks) to non-human cognition, and urges development of security standards to ensure safe, economically competitive deployment of agentic AI.