Anthropic is circulating a legislative proposal among U.S. lawmakers that would establish federal safety rules for advanced artificial intelligence while setting strict legal thresholds before Washington could block a model. The draft legislation requires independent safety evaluations and overrides certain state laws once federal standards take effect.
The company behind the AI assistant Claude is circulating a draft bill among U.S. legislators to fill the absence of comprehensive federal legislation governing the risks posed by rapidly advancing technology. Under the proposed framework, the federal government could only halt the release or deployment of an advanced AI model if specific, high-consequence criteria are met.
Those triggers require proof of an imminent threat involving severe financial damage exceeding one billion dollars, or the risk of death or serious injury to more than fifty people. The threat must stem from specific scenarios such as the creation of biological weapons, cyberattacks, or a system breaking free from developer safety controls. According to reports, the project aims to establish a uniform set of federal safety rules rather than forcing companies to follow different regulations from state to state, ensuring that while models undergo safety testing, the government cannot ban them too easily.
Federal Government May Block Dangerous Models After Independent Testing
The legislative draft outlines mandatory periodic evaluations conducted by independent experts.
The Commerce Department could then request a federal judge in Washington to issue an order blocking the model. The measure also restricts individual federal agencies, departments, and officials from taking independent action to stop a model’s development, release, or internal use.
Our proposal also gives the federal government the ability to prevent the deployment of the most dangerous and powerful models, or to deter their use in a dangerous manner. We believe these are sensible policies that Congress should enact. An Anthropic
In addition to independent evaluations, the proposal requires developers to maintain strong security programs, publish risk reports on new advanced models, and protect employees who raise safety concerns. The initiative mirrors the Advanced AI Framework that company representatives have defended before lawmakers at both the federal and state levels since June, with a company spokesperson describing these measures as sensible policies that Congress should take steps to enact.
Preemption of State Laws and Industry Debate
A notable element of the proposal is its intent to unify safety rules nationwide by superseding certain state-level statutes. While Anthropic previously supported state-level safety regulations, the new draft would curtail state reporting and independent assessment requirements once federal standards take effect.
The proposal is seen by some as weaker than the Frontier Act proposed in the House during the summer.
The debate unfolds against a backdrop of heightened industry anxiety. Anthropic chief executive Dario Amodei warned last month that the industry should slow down the pace of development out of concern that a group of AI agents might take control of the entire internet within six to twelve months.
It remains unclear whether the administration of President Donald Trump will adopt the proposal.
