Anthropic CEO Dario Amodei said poorly managed AI could threaten humanity, while OpenAI CEO Sam Altman warned people could lose control of the future to AI at a Sept. 23, 2026, U.N. Security Council session. They urged shared safety standards and capability benchmarks.
Altman proposed benchmarks measuring systems’ capabilities and whether company safeguards keep pace. Amodei called for shared international rules for AI testing and loss-of-control risk assessments, as well as agreements to prevent AI from assisting in biological-weapons development. He said Anthropic would slow deployment as much as needed for safety.
A European diplomat cited by Reuters said autonomous AI systems interacting could trigger armed conflict between states; U.N. Secretary-General António Guterres had warned about autonomous weapons the previous day. Governments still have no universally accepted definition of dangerous AI capability or widely recognized body with authority to evaluate frontier systems across borders, and what should happen after a failed safety evaluation remains unresolved.
