Anthropic’s AI Safety Bill Has an Alarming Threshold for Government Action

As the debate over regulating AI companies rages on, the artificial intelligence giant Anthropic has scaled up its Capitol Hill lobbying. And, as lawmakers propose ways to curb largely unregulated AI development, the company has started pitching its own draft bill to regulate the industry, Politico reported Friday.

The draft legislation would allow the government to block the release of any AI models it believes pose a danger to the public. But the its threshold for danger seems very high. “The bill would let the government block a model’s release or use only if it poses a foreseeable, imminent risk of more than $1 billion in damages, or of killing or seriously injuring more than 50 people,” Politico‘s Owen Dahlkamp reported. That is, if a model seemed like it might kill or seriously injure only a few people, or cause less than a billion dollars worth in property damage, the government would not be able to regulate it.

The bill further limits what counts as an intervention-worthy “risk” to three categories: the creation of a biological weapon, the model evading its developer’s controls, or a cyberattack. That means that other forms of AI-related harm—like AI-powered fraud—would not be included. Anthropic’s proposal also reportedly prohibits any US agency other than the Commerce Department from intervening to suspend an AI model’s development or internal use.

“This bill represents the proposals outlined in Anthropic’s Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June,” an Anthropic spokesperson told Politico. “These are sensible policies we believe Congress should act on.”

Lawmakers of both parties have increasingly called to regulate the AI industry—proposing to tax AI companies heavily, pause frontier model development, create a cabinet-level AI-oversight department, and more. Anthropic has ramped up its lobbying commensurately, spending more money lobbying Congress in the first half of 2026 than it did in all of 2025. The company has also lobbied heavily on the state level, particularly in California, where it has spent at least $90,000 to push pro-data-center-development policies in the past year. (Anthropic has also deployed lobbyists in states including Tennessee and North Carolina, and it has reportedly sent a delegation to Vatican City in order to convince the Pope to consider AI conscious.)

The company’s congressional draft bill represents an effort by the company to concentrate regulatory power in a single agency—and if it passes, it could preempt states from passing stronger regulatory laws.

That leaves AI regulation in the hands of a comparatively friendly federal government, led by a president who has suggested repeatedly that “artificial intelligence” be renamed “super intelligence.” He’s also pushed for industry self-regulation, which historically has not been very effective: see the 2007-08 financial collapse.

As Miranda Bogen of the Center for Democracy and Technology told my colleague Alex Nguyen in July, companies aren’t likely to change much without government intervention. “Without really robust interventions that actually change the incentives that companies face in what they’re testing…and what they’re doing to fix problems, I don’t think we’re going to address the risks that are coming online,” Bogen said.