Frontier AI safety could become a legal obligation for U.S. developers under a bipartisan Senate proposal.
The draft would require companies building models to reduce catastrophic risks and could allow federal authorities to block unsafe releases.
Senate Majority Leader John Thune, Commerce Committee Chairman Ted Cruz and Senator Amy Klobuchar are in talks, with Senator Maria Cantwell involved. Reuters reported that the measure would replace voluntary safety commitments with enforceable duties for frontier model developers.
From Voluntary Pledges to Federal Duty of Care
The proposal centers on a legal “duty of care” covering developers of frontier AI systems. Requirements under discussion include designing models to minimize catastrophic harm, allowing government intervention against unsafe releases and permitting court appeals of decisions.
The framework could involve national laboratories and government partners in testing advanced systems. Klobuchar told Reuters that her goal is “government oversight of the greatest risks posed by AI models.” She said developers should work with government experts to verify and test advanced models.
Cruz has described similar concerns. In an X post, he said he is working with Klobuchar and Thune to “address catastrophic risks involving biological or nuclear threats.” The discussions focus on leading developers rather than imposing identical requirements across the AI industry.
Lawmakers Respond to Rising Frontier AI Concerns
Reuters identified Google parent Alphabet, Anthropic and OpenAI among U.S. companies developing frontier AI models. The negotiations examine whether federal rules should override some state restrictions, reflecting calls for a consistent national framework.
The IAPP said the talks followed concerns raised by Anthropic researcher Jacob Coxon after his resignation. Coxon warned about competition to develop self-improving AI systems.
Senator Josh Hawley is investigating OpenAI’s role in the July Hugging Face cyberattack. Senator Bernie Sanders is working on legislation seeking to pause superintelligence development, according to the IAPP.
OpenAI has backed enforceable federal rules. In a September 9 policy article, Chief Global Affairs Officer Chris Lehane called for “mandatory capability-based national AI safety regulation.”
EU Rules Offer Existing Model for Oversight
The European Union already applies obligations to providers of general-purpose AI models presenting systemic risks. Those rules cover model evaluations, risk management, incident reporting and cybersecurity responsibilities.
Models exceeding 10^25 FLOP are presumed to carry systemic risk, although the European Commission can classify lower-compute models based on capabilities or effects.
Goldman Sachs Research expects global AI investment to reach $1 trillion in 2026, including $581 billion in the United States. Additional testing, legal and documentation requirements could affect release schedules and costs.
The bill’s timing remains uncertain. Reuters reported the House has one session week before November 3 midterms, while the Senate has three weeks.
