Anthropic has presented U.S. lawmakers with draft AI safety legislation, proposing a $1 billion damage threshold for government intervention and independent evaluations overseen by the Commerce Department. The bill aims to guide the rules covering advanced AI models.
TL;DR
- Anthropic proposes AI safety bill to U.S. lawmakers, setting a $1 billion damage threshold for government intervention.
- Bill calls for independent evaluations registered with the Commerce Department to review frontier AI models.
- Anthropic's proposal aligns with recent Senate bills on AI risk management and evaluation.
What happened
Anthropic, an AI lab, has been pitching U.S. lawmakers on draft artificial intelligence safety legislation, according to a report by Politico. The bill is based on Anthropic's Advanced AI Framework, which the company has been advocating for since June. The proposed legislation would allow the government to block an AI model's release or use if it poses a foreseeable, imminent risk of more than $1 billion in damages or of killing or seriously harming upwards of 50 people. The risks must be related to the creation or release of a biological weapon, an AI model evading its developer's controls, or a cyberattack.
The bill also calls for independent evaluations, registered with the Commerce Department, to periodically review the lab and its frontier models. If evaluators conclude there is an 'imminent enumerated risk' for a company's frontier models, they must report it to the Commerce Department within 72 hours of finishing their assessment, per the Politico report.
Why it matters
Anthropic's proposal is a notable effort by an AI lab to guide the rules covering its technology, potentially setting a precedent for AI safety legislation. The bill's focus on a $1 billion damage threshold and Commerce Department oversight could influence how other AI companies approach safety and regulation. The proposal aligns with recent Senate bills on AI risk management and evaluation, indicating a growing consensus on the need for AI safety measures.
Key facts
- Anthropic has been advocating for its Advanced AI Framework with federal and state lawmakers since June, according to the company.
- The proposed bill sets a $1 billion damage threshold for government intervention in AI model release or use.
- Independent evaluations registered with the Commerce Department would periodically review frontier AI models.
- Evaluators must report 'imminent enumerated risk' to the Commerce Department within 72 hours of assessment completion.
- Anthropic's proposal aligns with the Artificial Intelligence Risk Management and Security Act of 2026, introduced by Senate Democrats.
- Sen. Maria Cantwell's bill calls for the National Institute of Standards and Technology (NIST) to develop standards to protect against catastrophic harms from AI models.
- Sen. Josh Hawley's AI Risk Evaluation Act would place the Department of Energy in charge of stress-testing the most powerful AI systems before public release.
Context
Anthropic's proposal comes amid a broader push for AI safety and regulation. The White House, under the Trump administration, remains opposed to mandatory AI regulation, favoring a voluntary approach based on industry-developed standards. However, with the industry, the public, and many statehouses clamoring for action, White House opposition may not be enough to stop AI safety efforts. The proposal also aligns with recent Senate bills on AI risk management and evaluation, indicating a growing consensus on the need for AI safety measures.
