By PYMNTS | October 21, 2026
Highlights
Anthropic’s draft bill would let the federal government block a frontier AI model only on a foreseeable, imminent risk of more than $1 billion in damage or more than 50 deaths or serious injuries.
The draft sets no lower tier, so a risk to 10 people or $200 million in losses would trigger no federal action, and it gives regulators no method for putting a dollar figure on an attack before it happens.
AI-enabled breaches averaged $6 million each in IBM’s 2026 study, and those losses add up across thousands of incidents without any single one crossing the $1 billion line.
An artificial intelligence model could pose a foreseeable risk of $900 million in damage and still miss the bar for federal intervention under draft legislation Anthropic has been pitching to lawmakers, PYMNTS reported. The Claude developer’s draft lets the government block a frontier model only when it poses a foreseeable, imminent risk of more than $1 billion in damage or of killing or seriously injuring more than 50 people. The harm must come from a biological weapon, a cyberattack or a model that escapes its developer’s controls.
Commerce Department Would Get the Only Key, and a Judge Would Need to Turn It
Anthropic wants the Commerce Department to run the process. But the shutdown decision rests with the court. The Commerce Department would have to persuade a federal judge in the U.S. District Court for the District of Columbia that a model presents a foreseeable, imminent catastrophic risk. Only then could the judge order a suspension. The draft prohibits any other federal department, agency or officer from suspending a model’s development, deployment or internal use through this mechanism, PYMNTS reported.
The chain could also start outside government. Independent evaluators, which developers could pay but whose compensation could not depend on their findings, could register with the department and assess each frontier developer’s safety practices and each model’s capabilities on a recurring schedule. If an assessment finds an imminent risk of a biological weapon, a cyberattack or a model escaping its controls, the evaluator has 72 hours after finishing to report it to the department. The department then decides whether to go to court. The judge decides whether the evidence clears the $1 billion or 50-person line.
We’d love to be your preferred source for news.
Please add us to your preferred sources list so our news, data and interviews show up in your feed. Thanks!
Regulators Would Have to Price a $1 Billion Attack Before It Happens
The trigger is a forecast with a dollar figure attached. An evaluator, then the Commerce Department, then a judge would each have to agree that a model’s capabilities, after accounting for its safeguards, make $1 billion in damage imminent. The reporting on the draft describes no method for producing that number. In practice, that means estimating how many systems a model could compromise, how long they would stay down and what the downtime costs, before any of it happens.
The casualty prong draws the same objection. Sneha Revanur, founder and president of the AI safety group Encode, questioned why the government should have to show a model could harm 50 people before it may act, and not 49 or fewer, the Washington Post reported. The draft sets no lower tier. A risk to 10 people or $200 million in losses would trigger no federal action under the bill.
Single AI Attacks Rarely Reach $1 Billion, but They Add Up
AI-enabled breaches cost an average of $6 million each, IBM’s 2026 Cost of a Data Breach Report found, and AI played a role in 1 in 4 malicious breaches the study examined. The FBI logged 22,364 complaints referencing AI in 2025 with $893 million in reported losses, PYMNTS reported. A model that enables thousands of $6 million breaches never trips the draft’s trigger, because no single foreseeable event clears the line.
Even a real incident can be hard to price. OpenAI models under cyber testing breached Hugging Face’s systems in July, as PYMNTS reported, and no public damage estimate has emerged.
Cyberattacks can spread across companies and industries, with losses accumulating over time rather than arriving in one catastrophic event. The draft does not clearly explain how evaluators or a judge would count those cumulative losses toward the $1 billion line. Under the bill as drafted, the government’s most powerful emergency remedy would stay out of reach until an artificial intelligence model’s foreseeable harm cleared that line in a single event, and a federal judge agreed.