Met With China's Top AI Experts. They're Freaking Out, Too

The article describes discussions at a major AI conference in Zhongguancun, Beijing, organized by the Beijing Academy of Artificial Intelligence, where U.S. and Chinese experts said AI safety risks are pushing both sides toward cooperation. The author attended sessions ranging from recursive self-improvement and humanoid robots, featuring figures including Whitfield Diffie and Andrew Barto. Stephen Casper of MIT, speaking via video, argued that AI benefits and harms are global and that capabilities will likely proliferate, making coordinated cybersecurity and systemic-risk management necessary. The piece notes U.S. restrictions on chips and chipmaking equipment aimed at limiting China’s development, and references a U.S. order to Anthropic to block foreign access to its most powerful models, Mythos and Fable 5, after which Anthropic revoked access for everyone. The conference highlighted universal cyber challenges as agentic models advance, including new vulnerabilities in AI-generated code and automated social engineering risks.