Report: Anthropic Draft Sets $1B Damage or 50-Casualty Threshold
Anthropic has reportedly circulated a draft federal bill under which the government could block an AI model only over a risk of $1 billion in damage or more than 50 deaths or serious injuries. It is a private draft, not a public proposal.
Politico reported on Oct. 16 that it had obtained a confidential draft federal AI bill that Anthropic circulated among lawmakers. It cited three people familiar with the matter. An Anthropic spokesperson told Politico these are “sensible policies we believe Congress should act on.” We could not read the Politico story or the draft text itself, so this post relies on a Benzinga article that cites Politico. The threshold applies only to blocking a model; developers’ disclosure, evaluation, and reporting duties are separate.
What the $1 billion and 50 people measure
Under the draft, the government can stop the release or use of a frontier AI model, meaning one of the most advanced AI models, only when it poses an imminent risk. The threshold is at least $1 billion in damage or more than 50 deaths or serious injuries. The named sources of risk are a biological weapon, a model slipping out of its developer’s control, and a cyberattack.
This is often retold as “the line where regulation starts,” but what was reported is closer to the threshold for the government’s last resort. The process to use it is demanding. The Commerce Department has to ask, and only a federal judge in the U.S. District Court for the District of Columbia can order a suspension. Other agencies cannot halt a model’s development or use.
Evaluation comes first. Independent evaluators registered with Commerce would test frontier models. Developers may pay them, but the fee cannot depend on the result. An evaluator that finds an imminent risk must report it within 72 hours. The report says this part resembles California’s AI laws, such as SB 53.
How far state laws would be replaced
Reports agree that the draft would have federal law replace some state AI laws. Benzinga names risk disclosure, outside evaluation, and incident reporting, but its article says “four areas” and lists only three, so we could not pin down the exact scope. Other outlets describe it as broader.
Some coverage sees tension with a policy document Anthropic published itself. That document says Congress should not preempt state law unless the federal law is at least as strong as the framework it proposes, and that any preemption should be narrow enough to leave areas like child safety and consumer protection to the states. Benzinga says the draft is tied to Anthropic’s Advanced AI Framework, but we have not checked the draft against it, and we do not know how its replacement clause meets that condition.
Congress already has competing efforts. In the House, the FRONTIER Act from Reps. Lori Trahan and Jay Obernolte would let the government stop deployment of risky models. In the Senate, John Thune, Amy Klobuchar, and Ted Cruz are negotiating a bill on catastrophic AI risk. Democrats have resisted replacing state laws.
What to watch next
- Whether the draft is introduced as a bill, and whether its text becomes public and settles the scope of state-law replacement
- Whether the House FRONTIER Act and the Senate negotiations adopt this threshold
- Whether other AI companies put out similar drafts or positions
Sources
Source: www.benzinga.com