Anthropic pitches AI safety bill to US Congress

Anthropic’s draft bill aims to give the federal government a shutdown tool

Anthropic told lawmakers on Capitol Hill that the United States should be able to stop a model’s release if the model is likely to kill or seriously injure more than 50 people, or cause over $1 billion in damage. The thresholds – drawn from a Politico report on 16 October 2026 – would apply to risks such as the creation of biological weapons, an AI system escaping its developer’s control, or a large‑scale cyber‑attack.

How the bill would work in practice

The draft requires developers of “frontier” AI models to run safety tests, publish the results and allow independent assessors to review the models on a regular basis. Those assessors would register with the U.S. Department of Commerce and, crucially, could not be paid in a way that ties compensation to a favourable outcome. If an evaluator flags an imminent risk, they must alert the Commerce Department within 72 hours. The department could then ask a federal judge in the District of Columbia to issue an injunction halting the model’s deployment.

Anthropic’s rationale and industry reaction

An Anthropic spokesperson told AFP that the proposal “grants the federal government the ability to block or deter dangerous deployments of the most powerful models.” The company says the framework would complement, not replace, voluntary industry safeguards and would give the government a backstop for truly catastrophic scenarios.

Industry commentator Charlie Bullock of the Institute for Law & AI called the plan “pretty reasonable,” though he noted on X that it is weaker than the competing FRONTIER Act introduced by Rep. Lori Trahan and Rep. Jay Obernolte.

The bill in the broader congressional landscape

Anthropic’s effort joins two other high‑profile proposals:

• AI Kill Switch Act – introduced by Rep. Nathaniel Moran (R‑TX) and Rep. Ted Lieu (D‑CA). It would trigger safeguards when a model threatens $100 million in economic damage or 10 deaths.
• FRONTIER Act – championed by Rep. Lori Trahan (D‑MA) and Rep. Jay Obernolte (R‑CA). It calls for nationwide transparency, incident reporting and a licensing regime for independent auditors.

Both of those bills set lower thresholds than Anthropic’s draft, reflecting a split in Congress over how aggressive federal intervention should be.

Recent friction between Anthropic and the U.S. government

Anthropic’s push comes after the firm was banned from Pentagon contracts in February 2026 and was forced in June to suspend its most advanced models for nearly three weeks. CEO Dario Amodei warned in a June interview that a group of autonomous AI agents could “take control of the entire internet” within six to twelve months, underscoring the urgency the company says it feels.

What’s next?

The draft has not yet been formally introduced as legislation. Anthropic is urging Congress to act before the mid‑term elections, when the political calculus around tech regulation could shift dramatically. If a judge were to issue an injunction under the proposed framework, it would be the first time a federal court directly blocks the deployment of an AI system.

---

Comparison of three AI‑risk bills

The table shows Anthropic’s higher thresholds but a clearer federal enforcement path compared with the other two bills.