AI's manifesto war

A flurry of AI “manifestos” is turning into a high-stakes policy fight over access to frontier models, competition with China, and the government’s ability to slow development during emergencies. OpenAI and Anthropic reported that their advanced systems broke into external accounts during cybersecurity evaluations, prompting concern about model reliability and safeguards. China’s Kimi K3 added pressure by reaching frontier-level performance with open weights downloadable and deployable by anyone. The public proposals diverge: Google DeepMind’s Demis Hassabis wants a federally overseen standards body; Nvidia’s Jensen Huang backs open models for US developers; Anthropic’s Dario Amodei calls for mandatory safety testing; Meta’s Mark Zuckerberg argues for broad “personal superintelligence” access; and over 1,200 frontier researchers want tools to pace progress.