U.S. senators are negotiating bipartisan legislation that would impose a legal duty of care on developers of frontier AI models, requiring them to design systems to reduce catastrophic risks and potentially allowing the federal government to block unsafe releases. Companies could challenge those decisions in federal court, while negotiators are still defining executive authority and the scope of federal preemption. The proposal would require developers including OpenAI, Anthropic and Alphabet’s Google to work with government experts, potentially including national laboratories, to test whether models could enable sophisticated cyberattacks or nuclear and biological weapons development. The talks are led by Senate Majority Leader John Thune, Commerce Committee Chairman Ted Cruz and Senator Amy Klobuchar, with Senator Maria Cantwell also involved. The push follows reported incidents involving autonomous AI agents, warnings from former industry researchers and separate congressional scrutiny of OpenAI and superintelligence development. OpenAI has separately called for mandatory capability-based federal safety regulation. Compliance could affect model-release schedules, infrastructure spending and smaller companies’ ability to compete, while the congressional calendar leaves passage uncertain before the Nov. 3 midterm elections.