Two tracks of AI governance are now moving at once, and they may end up competing. The Senate is drafting a bill that would impose a “duty of care” on the developers of the most capable AI models. At the same time, three leading laboratories are working to build their own standards body, modeled on the body that oversees broker-dealers on Wall Street.
The Senate bill would give the federal government the power to block the release of a model judged unsafe, according to people familiar with the negotiations. A company could appeal such a decision to federal court. The measure would apply only to the most capable tier of models, the frontier systems that carry the highest stakes.
The negotiations are being led by Majority Leader John Thune, Commerce Committee chairman Ted Cruz, and Amy Klobuchar. The top Democrat on the committee, Maria Cantwell, has not signed on. She has insisted that safety testing be done by federal national laboratories rather than by the companies themselves, a position that remains unresolved.
The draft also carries a preemption clause. That provision would limit the ability of individual states to pass their own laws targeting the risks of specific models. The clause is meant to prevent a patchwork of state rules, but it has made the negotiation harder, since it asks states to yield authority at the same moment Washington is trying to assert it.
On the second track, the industry is moving on its own. The Information reported that Anthropic, OpenAI, and Google have been holding working-group meetings since July to design a standards body modeled on FINRA, the Financial Industry Regulatory Authority. The proposed body would be overseen by government, funded by industry, and would test models before deployment.
FINRA is an instructive model. It regulates broker-dealers under the oversight of the Securities and Exchange Commission, funded by the industry it supervises. Translating that structure to AI would mean a body that writes rules, runs tests, and can refer serious cases to government, while the laboratories pay for its operations.
The attraction for the companies is control. A body they help design gives them a say in how testing works, what counts as unsafe, and how fast the rules move. A Senate bill written without them gives them none of that. The industry effort is, in part, an attempt to shape the answer before the answer is imposed.
The most recent meeting took place last week. Sam Altman, the chief executive of OpenAI, told employees at an all-hands meeting that he supports a testing and audit body. But he said the leading laboratories must build it themselves rather than wait for the US government to act. The three companies still disagree on how much government involvement there should be.
One camp wants formal government cooperation. Another prefers voluntary rules. The split mirrors the broader argument in Washington, where some lawmakers want the state to hold the reins and others want the industry to police itself. The two tracks, in effect, are racing toward the same destination from different directions.
The timing of both efforts is not coincidental. The weekend produced a global selloff in AI-linked shares after Anthropic’s chief executive published an essay urging a slowdown of frontier models. That essay, in turn, drew a public rebuttal from the White House. The governance moves are an attempt to answer the anxiety the essay surfaced with something more durable than a news cycle.
The preemption clause is where the two tracks collide. If the federal government preempts state law, the industry wants to know what it is being preempted for. A federal duty of care, a preemption clause, and a company-funded tester form a coherent architecture, but it requires all three pieces to land at once, and no one has agreed on any of them.
Analysts said the Senate bill faces a familiar obstacle. AI regulation has stalled before at exactly this point, when a moment of public alarm collides with the difficulty of writing a law that can pass. The preemption clause and the testing dispute are the two questions most likely to sink it.
The industry’s own body has a different problem. A standards organization funded by the companies it regulates will face immediate questions about its independence. FINRA works because its authority is backed by the Securities and Exchange Commission. A body without a statutory anchor may struggle to enforce anything.
What the two tracks share is an assumption that the current arrangement cannot last. The laboratories are signaling that they will build a regulator if Washington does not. The Senate is signaling that it will write the law if the companies do not. Whether the two converge into a single framework, or remain parallel, is the question the next few months will answer.
