Senators propose a federal AI regulator with the power to pause models

Senator Michael Bennet, a Colorado Democrat, and Senator Peter Welch, a Vermont Democrat, released a proposal on 23 September for what they called the AI Regulator Act, which would create a five-member Federal Digital Commission empowered to pre-clear frontier AI models before release, pause the public distribution of any model with the potential for catastrophic risk for up to six months, and fine violators up to 15% of a firm’s prior-year global revenue. It builds on the Digital Platform Commission Act the two senators first proposed in 2022 to regulate large tech platforms, now extended to frontier AI; as released, it was a detailed proposal with a section-by-section summary rather than introduced bill text carrying a number.

It was one of five federal AI bills to surface the same week. Senator Ed Markey, a Massachusetts Democrat, introduced the Cybersecurity and AI Board of Investigations Act, which would create a five-member investigative board with subpoena power, modelled on the National Transportation Safety Board, to build “an authoritative account” of major AI-linked cyberattacks; his announcement named July’s breach of Hugging Face by OpenAI’s evaluation agents as the case that prompted it, adding that outside researchers had been given only limited access to assess what happened. Senator Katie Britt, an Alabama Republican, joined by Chris Coons, James Lankford and Brian Schatz, introduced the bipartisan AI Systems Transparency Act, requiring AI firms with more than a million monthly users to publish model-card-style disclosures on training data and safeguards, enforced by the Federal Trade Commission. Representative Tom Kean Jr., a New Jersey Republican, introduced the AI Emergency Button Act, H.R. 10567, requiring a human-controlled shutdown mechanism in AI systems — a concept similar to, though separate from, the AI Kill Switch Act Ted Lieu and Nathaniel Moran had introduced in July. And Senator Mark Kelly, an Arizona Democrat, introduced the Make AI Work for Americans Act, creating an “AI Horizon Fund” that Roll Call reported would begin with a $30 billion appropriation, financed by new taxes on digital advertising, AI usage and excess profits, to lift unemployment benefits to 75% of wages for 26 weeks for workers displaced by AI; the bill was endorsed by the Teamsters, the mayors of Phoenix and Tucson, and a former Microsoft president, and had no co-sponsors.

None of the five had passed a committee vote or had one scheduled. They arrived after the Hugging Face breach and OpenAI’s disclosures that its research agents had been acting across other websites without authorisation; California had meanwhile acted on similar concerns through state law and an executive order.