Five moves in AI politics this week

Over the past two weeks the AI debate moved from the labs into politics. The question is simple: should the development of frontier models be deliberately slowed, or should the race go on?

This week answers came from five places: the Trump-Xi summit in Washington, Meta CEO Mark Zuckerberg, two Democrats in Congress, a memo from Trump’s circle and the New York City Council. The debate reached this point in several steps.

On September 12 Anthropic CEO Dario Amodei published “We Must Pace the Frontier,” calling on the industry to deliberately slow how fast model capabilities improve. Sam Altman and Elon Musk endorsed it within hours. On September 18 four consumers sued the labs for antitrust, treating the call as an agreement among competitors.

On September 19 Trump announced an AI Force and an AI czar, framed as protecting the industry. On September 23 Amodei and Altman asked the UN Security Council for global safety standards, and Amodei said AI could be a risk to humanity as a whole.

During Xi Jinping’s visit to Washington on September 23-25 the two countries reached an eight-point consensus. On AI the concrete outcome was a channel: a dialogue on superintelligence and a communication line for AI-related incidents, with the first round expected in November.

No rules or limits were set. Analysts say neither side wants to slow the race. The summit’s other headline result was reciprocal tariff cuts on $30 billion of goods each.

Mark Zuckerberg told NBC News on September 24 that the industry does not need a coordinated slowdown: “I don’t think that we need some kind of industrywide coordination.” In his view each lab should take time internally when it sees issues; he pointed to Meta delaying its Muse agent for safety.

It is an open break with the joint action Amodei and Altman are calling for.

Senator Bernie Sanders and Representative Greg Casar introduced the Ban Artificial Superintelligence Act on September 23. It permanently bans developing or deploying AI that exceeds human cognitive performance across most domains or could destroy or disempower humanity, and pauses advanced AI development until a federal regulator sets safety rules.

A new cabinet-level Department of Artificial Intelligence would enforce it. Violating companies face the corporate death penalty and individuals up to 20 years in prison. It has long odds in the Republican-controlled Congress.

An opposition research memo dated September 22, published by Axios on September 24, targets Anthropic and its CEO Dario Amodei. It was written by a Trump political adviser who does not work for the White House, and circulated among White House officials this week.

The memo ties Anthropic to effective altruism, titles one section “how EA built the AI-doom pipeline” and calls the Amodei siblings “The Anthropic knot.” Both have previously said they are not part of the movement.

New York City Council Speaker Julie Menin unveiled a 10-bill package on September 25. Any AI system sold in the city would need independent validation and a kill switch, with a $25,000 fine per violation.

Whistleblowers would get a share of the fines, and people harmed by jailbroken tools could sue. The CEOs of the major labs have been called to an October 5 hearing.

The week splits into two camps. On one side are those arguing for slowing down or setting limits: Amodei, Sanders and Casar, and the New York City Council. On the other are those who do not want to slow the race: the Trump administration, Beijing and Zuckerberg.

Washington still has no federal AI safety law. States and cities are filling the gap for now: under New York State’s RAISE Act, major frontier developers must register with the state from November and, from January 2027, report critical safety incidents within 72 hours.