Anthropic Pitches Draft AI Safety Bill to Lawmakers
Anthropic has been quietly shopping its own AI safety bill around Washington. Politico first reported the draft on October 17, 2026, and Reuters confirmed the details. The proposal would let the US government ask a court to halt a frontier AI model, but only in narrow, catastrophic scenarios.
This is the AI lab writing rules for itself. And the thresholds it picked are telling.
What the bill would do
The draft sets an extremely high bar for intervention. The government could only seek a halt if a model’s risk exceeds $1 billion in damages or threatens to kill or seriously injure more than 50 people. The risk must come from one of three categories: bioweapons development, a cyberattack, or the model escaping its developer’s controls.
Here is how enforcement would work. Independent evaluators would register with the Commerce Department and periodically assess frontier labs and their models. Labs could pay these evaluators, as long as payment is not tied to the outcome. If an evaluator finds an “imminent enumerated risk,” it must report to Commerce within 72 hours. Commerce could then ask a federal judge in Washington D.C. to order a suspension.
One striking provision: the bill explicitly bars every other government agency from suspending a model’s development or deployment on its own. All enforcement power would sit with Commerce and the courts.
The draft also requires labs to publish safety frameworks, issue transparency reports, and protect whistleblowers. Anthropic says this matches its Advanced AI Framework published earlier this year.
The controversial parts
Two things will draw fire.
First, the thresholds. $1 billion and 50 lives are enormous bars. Safety advocates will argue this leaves serious but smaller-scale harms completely unaddressed at the federal level. Bias, misinformation, and labor displacement are not even in the bill’s risk categories.
Second, preemption. The draft would override some state AI laws on risk disclosure and incident reporting. That puts Anthropic on a collision course with state attorneys general, which is awkward timing given eight states just sued OpenAI this same week. Democrats have already resisted similar preemption proposals, Politico noted.
There is also the obvious conflict. The companies being regulated wrote the draft. An Anthropic spokesperson called the policies “sensible,” which is exactly what you would expect the author to say.
Where this fits
This is not happening in a vacuum. The regulatory scrutiny of AI labs keeps intensifying. The EU AI Office and the FTC both opened files on OpenAI this week over agent safety failures. Congress has competing proposals, including the House FRONTIER Act and a Senate bill from Thune, Klobuchar, and Cruz.
Anthropic’s move is strategic. By proposing its own framework, it tries to shape the rules before someone else writes harsher ones. Whether lawmakers accept a bill drafted by the industry it regulates is another question. No sponsor has been named. No formal introduction has happened.
As advanced AI systems keep improving, the fight over who sets the guardrails is becoming one of the defining AI trends shaping 2026. This draft is Anthropic’s bid to hold the pen.
FAQs
What does Anthropic’s draft bill propose?
It would let the Commerce Department ask a federal judge to halt a frontier AI model, but only for imminent catastrophic risks: over $1 billion in damages, more than 50 deaths or serious injuries, from bioweapons, cyberattacks, or loss of control.
Has the bill been introduced in Congress?
No. It is a draft pitched to lawmakers. No sponsor has been named and no formal introduction has occurred.
Why is the bill controversial?
Critics point to the extremely high intervention thresholds, the exclusion of harms like bias and misinformation, and provisions that would override state AI laws. The fact that AI labs drafted their own regulation also raises conflict-of-interest concerns.
