Anthropic proposes AI safety bill with strict intervention thresholds - Politico

Anthropic, an artificial intelligence startup, has been quietly presenting a draft bill to Capitol Hill lawmakers, according to a report by Politico on Friday. The bill aims to regulate the safety and security of frontier AI models, according to three sources familiar with the efforts. The proposed legislation allows the government to block or deter the release or use of a model if it poses an imminent risk of causing more than $1 billion in damages or resulting in the death or serious injury of more than 50 people.

This risk could stem from the creation or release of a biological weapon, the model evading its developer's controls, or a cyberattack.

Anthropic's spokesperson confirmed that the bill reflects proposals outlined in the company's Advanced AI Framework released earlier this year. The legislation grants the government the power to block or deter the deployment of powerful models, requiring developers to test such models, be transparent about findings, and undergo independent evaluations.

Independent evaluators, registered with the Department of Commerce, would periodically assess the labs and their frontier models. These labs could pay these evaluators, as long as the payment is not contingent on the evaluation's outcome. If the evaluators determine there is "imminent enumerated risk" for a company's frontier models, they must report it to the Commerce Department within 72 hours of completing their assessment.

The Commerce Department can request a federal judge to halt a model if it determines there is risk, though this prohibition extends to any other US government department, agency, or officer from suspending development, deployment, or internal use.