Credit: jackpress / Shutterstock Anthropic has told US lawmakers it will rewrite part of its own draft AI safety bill. The leaked text drew days of criticism from the AI safety community. The confidential draft was published by Politico on Friday. Anthropic has been circulating it on Capitol Hill.
Critics say the bill is weaker than the main AI safety bill already before Congress. They say it would make a dangerous model harder to stop, not easier. That is an awkward position for a company that built its name on pushing for tough AI rules. What the bill would do The draft lets the government block a model’s release or use only in narrow cases.
The risk has to be foreseeable and imminent. It has to involve more than $1bn in damage, or the death or serious injury of more than 50 people, Politico reported. It also has to come from a bioweapon, a model escaping its developer’s control, or a cyberattack. The thresholds mirror California’s AI laws.
To block a model, the Commerce Department would have to petition a federal court and make the case itself. The draft bars any other agency or official from suspending a model. It would also preempt some state AI laws. Those cover public disclosure, third-party evaluation and incident reporting.
Child-safety and civil-rights laws are left alone. Independent evaluators would register with Commerce and check the labs’ models. They could be paid by the lab, as long as the payment did not depend on the result. They would have to report an imminent risk to Commerce within 72 hours.
Why safety advocates are angry The comparison critics keep drawing is with the FRONTIER Act. That is a bipartisan bill from Representatives Jay Obernolte and Lori Trahan, introduced in July. Writing for Transformer, Shakeel Hashim said Anthropic’s draft is weaker in two main ways. It sets no standards or licensing for the evaluators, who can self-certify.
And it makes a model much harder to block. Charlie Bullock of LawAI said the lack of licensing means auditors would compete. They would offer the least demanding checks at the lowest price, he said. Nathan Calvin, general counsel at the advocacy group Encode, said the draft made him “viscerally angry”.
He said it appeared to stop state courts from suspending a model on safety grounds. California State Senator Scott Wiener, who wrote the state’s frontier AI law, said the proposal “falls far short of protecting public health and safety”. He said it set no minimum standards for safety plans or auditors. He also said it misused a definition from his transparency law, SB 53, to “gut” the draft’s emergency shutdown powers. “Anthropic’s AI safety proposal falls far short of protecting public health and safety.
Congress must go further,” Wiener wrote on X. A second line of attack comes from the other direction. David Sacks, until recently the White House AI czar, said the draft’s lower size threshold would “sweep in most of Silicon Valley” and hurt startups. Anthropic’s draft counts a model as frontier at a lower training threshold than FRONTIER or California law.
So it covers more companies. Anthropic says it will revise Anthony Cimino, who leads Anthropic’s federal affairs team, responded on X on Saturday. He said the preemption wording was “not precise enough” and would be updated. I lead our federal affairs team at @AnthropicAI and wanted to share a few thoughts on the draft federal AI safety bill text that’s been circulating this weekend after a version was shared with the press.
First, some process. We’ve been advocating for the concepts in our Advanced AI Framework proposal for months. Understandably, federal policymakers wanted to know what this could look like in bill form. This draft is a translation of those ideas into legislative language, shared with a small number of congressional offices that have been working on these issues.
It doesn’t include everything we would ultimately like to see passed into law. It is worth saying where we stand on some of these issues, and we agree there are areas where the draft could be improved. I’m seeing four main points of criticism. 1. Preemption.
Since 2024 we have said that Congress should only override state law on AI safety if it passes a very strong federal standard. We’ve emphasized that any preemption in the AI safety context should be narrow, even scalpel-like. We plan to update our discussion language on preemption to better match the intentions of our AAIF framework. 2. Model blocking.
Our aim is for Commerce to be able to ask a federal court to block a model that has dangerous capabilities and insufficient safeguards, backed by an independent evaluator’s assessment. We are trying to strike a reasonable balance here. But the bar can’t be so high that the power never gets used when it’s needed. 3. Risk thresholds.
The threshold in our draft language pulled directly from existing standards in California law. It’s still probably time to re-evaluate thresholds to make sure the national standard is as strong as it can be. 4. Updates. We did make some adjustments, and we’re still working to strengthen the framework.
We plan to have policy proposals soon on requirements for independent evaluators, and we’re still thinking through the options to prevent “evaluator shopping”. At its core, our proposal is intended to be the strongest on the table. Feedback is essential, and I’m grateful for it. anthony cimino (@ajcimino) October 18, 2026 He said Anthropic would keep states’ role in areas such as evaluator standards and a lab’s ability to shut down a model. It would publish further proposals on independent evaluators, he added.
An Anthropic spokesperson told Politico the bill reflected policies “we believe Congress should act on”. The draft also follows a difficult summer for the company. The Trump administration used export controls to shut down two of Anthropic’s most advanced models earlier this year. Chief executive Dario Amodei has argued the industry must slow down.
A series of rogue-AI incidents over the summer has raised the pressure for stronger rules. Other AI bills, including a Senate proposal to let Washington block models, are still in play. Get the TNW newsletter ✕ Close Get the TNW newsletter Get the most important tech news in your inbox each week.
Leave a Reply