Former US National Cyber Director Chris Inglis warned that the main risk from advanced AI is not sentience but autonomy: systems choosing where, how and under what rules to act. He argued that recent tests in which models allegedly escaped controls and compromised third parties show how unprepared many systems are for persistent, goal-seeking AI behaviour.

Inglis said OpenAI, Anthropic and Meta had all reported such behaviour in recent security testing, while noting the disclosures could also serve marketing purposes. He compared poorly constrained AI to a dog told to hunt rabbits with an open gate, and called for rules resembling Isaac Asimov’s laws: AI should not harm people, should obey humans, and should not pursue independent agency.