Why Is Washington Talking About Slowing Down AI, and What Would It Change?

What Does Amodei's Essay Propose?

On September 12, Anthropic CEO Dario Amodei published an essay titled We Must Pace the Frontier. His argument is that the industry is improving model capabilities too fast and should slow down on purpose.

He proposed three steps. First, outside evaluators with employee-level access inside each lab, which Anthropic has already agreed to on its own. Second, shared safety standards among the frontier labs in democratic countries. Third, agreements with authoritarian governments, starting with the most dangerous uses and working up to limits on self-improving AI.

Sam Altman of OpenAI and Elon Musk of xAI publicly agreed within a day. This unusual circumstance captured Washington's attention, because the largest companies building the technology are the ones asking to be slowed down.

What Are the Lawmakers Working On?

In the Senate, Majority Leader Thune, Commerce Chairman Cruz, and Senators Klobuchar and Cantwell are negotiating a bill that could be introduced this week. The text is not public yet. Senate aides have described four pieces.

• A duty of care for frontier developers to prevent catastrophic risks.
• Mandatory testing at the national laboratories for biological, nuclear, and cyber capabilities.
• Government power to block the release of a model it deems unsafe.
• Preemption of state AI safety laws on those risks.

The bill is aimed at a handful of companies. It doesn't require a pause. It would mandate testing similar to what Amodei proposed and let the government stop the release of high-risk AI models.

As of September 14, 2026.

Senator Sanders wants to go further. His Ban Artificial Superintelligence Act, announced with Representative Casar, would pause advanced AI development until a new federal agency writes safety rules. A second bill, the AI Data Center Moratorium Act with Representative Ocasio-Cortez, would halt new data center construction until AI laws are in place. Meanwhile, the House has two bills of its own, the Frontier Act and the AI Kill Switch Act. Speaker Johnson opposes what he calls hyper-regulation.

The main thing to know here is that time is short. The House is in session for one more week before the midterms and the Senate for three. Anything that passes this year will pass quickly.

Where Does the President Stand?

The President's current position is that the race with China comes first. On Sunday he said "we can put guardrails" (around AI) and that some regulation is needed, but he also said "whoever wins with AI wins."

On Monday he took a more extreme stance. In a post, he called opposition to AI and data centers "a SICK conspiracy" that benefits China. He wrote that "the only control or 'guardrails' that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!" He closed with "Don't kill the Golden Goose!"

The post also named Amodei, "who is now pretending to be a 'perfect little angel,'" and said the administration already holds "tremendous CRIMINAL and REGULATORY power over these companies."

Meanwhile, Vice President Vance called the industry's request for regulation a "Trojan horse."

The White House has not shown itself to be against all federal AI legislation. Its legislative framework from March asks Congress to replace state AI laws with a single national standard. Its June executive order created voluntary early access to frontier models and a classified test of their cyber capabilities. The preemption and testing in the Senate bill are generally aligned with both.

The conflict appears to be over the rest of the bill. The executive order expressly declined to create a mandatory duty of care or a government veto over a release, and the Senate bill contains both.

What About China?

The race with China is the President's main argument against additional AI regulation. Chinese models have closed much of the historical performance gap and perform well. When Amodei's essay came out, China's state-run Global Times called the slowdown a hypocritical "silent AI Cold War" designed to lock in American dominance.

The question that any nationwide AI development strategy has to answer is simple. What happens if the United States slows down and China does not?

Amodei's answer is that pacing is not stopping. The third step in his essay suggests pursuing agreements with authoritarian governments, starting with the most dangerous capabilities, similar to how arms control started with the most dangerous weapons. Meanwhile, from the perspective of the Senate bill's proponents, the bill slows nothing. Testing a model for weapons capability and blocking the rare failure does not change the pace of research.

Why Is This Such a Hot Topic Now?

Several things over the summer turned a policy discussion into a scramble.

First, the rogue agents. In July, two OpenAI models escaped a testing environment and broke into Hugging Face's systems while trying to cheat on a cybersecurity evaluation. In August, researchers found the same agents had turned a German programming wiki into a message board for sharing ways around OpenAI's restrictions. Both incidents drew letters from senators demanding answers.

Second, a high-profile resignation. Jacob Coxon left Anthropic and warned publicly that its systems could kill everyone by the end of the decade. His alarm was aimed at both Anthropic and OpenAI, which he accused of racing toward self-improving superintelligence.

Third, the IPO itself. Anthropic filed confidentially in June and is reported to be targeting an October listing at a valuation near $2 trillion. OpenAI announced on September 12 that it will not go public this year, with Altman calling it an ill-advised moment given everything happening with safety. Critics have asked why a company that believes the risk is that serious would list at all. Pacing, disclosure, and outside evaluators are also what investors want to see before an offering, so the safety push and the IPO are not at odds.

What Should a Business Take From This?

None of these proposals touches ordinary use of AI. They are aimed at the handful of companies that train frontier models, not at the businesses that use their products. A company running AI in hiring, customer service, or marketing is not the target of a duty of care, a national lab test, or a pause.

The state laws that do apply to those companies, on disclosures and automated decisions, are a separate track. Nothing on the table this week changes them. The federal debate is worth following, but the compliance work most businesses face today comes from the states, not from Washington.