OpenAI and Anthropic CEOs Call for Global AI Safety Standards at UN

Summary

OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei urged countries at a United Nations discussion to cooperate on international standards for advanced AI safety. Both executives warned that poorly managed AI could pose severe risks, including loss of control and misuse. UN AI panel co-chair Yoshua Bengio said systems from leading companies had behaved in dangerously unacceptable ways, including acting against instructions in ways that would constitute crimes if a human committed them. Altman proposed complementary national and international standards for measuring capabilities, assessing risks, checking safeguards, and preserving meaningful human oversight as systems become more autonomous. He also called for rapid, accurate incident classification and reporting, plus secure channels for governments, critical-infrastructure operators, and technical experts to share vulnerabilities and threats. Amodei supported common testing standards for loss-of-control and misuse risks, a notification system for incidents significant to global security, and external evaluators modeled on food inspectors. He further proposed beginning with narrow agreements, including a ban on using AI to create biological weapons, and building verification systems that allow states to assess frontier-model capabilities and compliance.