Washington's AI debate is moving from voluntary commitments toward potential legal duties. Reuters reports that bipartisan Senate negotiators are considering legislation that would require developers of the most advanced AI systems to take reasonable steps to mitigate known major risks.
The proposal is aimed at frontier-model developers rather than ordinary software companies. Risks under discussion include models materially assisting nuclear or biological weapons development, major cyber misuse and other forms of catastrophic harm.
A federal stop-release power would be a major shift
One of the most consequential ideas is federal authority to prevent release of an AI system judged unsafe. Companies would retain the ability to challenge such a decision in federal court, but the existence of a pre-release intervention mechanism would materially change the compliance landscape for firms such as OpenAI, Anthropic and Google.
Negotiators are also discussing testing at national laboratories, which could create a formal government evaluation channel for systems whose capabilities exceed ordinary commercial benchmarking.
Our view: capability thresholds will decide whether the law is workable
American Commerce Review's view is that the critical design question is scope. A rule aimed too broadly at AI software would be difficult to administer and could burden ordinary enterprise adoption. A rule tied to measurable frontier capabilities would be more targeted but would require regulators to update thresholds as models improve.
The next steps are legislative text, committee agreement and clarity on how federal rules would interact with California and other state regimes. With Congress approaching the November elections, timing remains uncertain.
| Possible requirement | What it would do | Open question |
|---|---|---|
| Duty of care | Require mitigation of known major risks | How negligence would be defined |
| Federal stop-release authority | Allow intervention before unsafe deployment | Threshold for intervention |
| National-lab testing | Provide government capability evaluation | Which models qualify |
| State-law preemption | Create more uniform national rules | How much state authority remains |
Frequently asked questions
What is the proposed AI duty of care?
It would require covered frontier AI developers to take steps to mitigate known major risks from their systems.
Could the US government block an AI model release?
That authority is reportedly under discussion, with companies able to challenge decisions in federal court.
Is the bill final?
No. Reuters reports that bipartisan senators are still negotiating the proposal.