Anthropic Proposes Draft Bill for Regulating Frontier AI Risks

A draft bill aimed at regulating the safety and security of advanced artificial intelligence models has been privately pitched to US lawmakers by Anthropic, a company that develops AI technology. The proposed legislation would give the government authority to block the release or use of an AI model if it poses a foreseeable and imminent risk of causing significant harm. According to Politico, which first reported on the draft bill, this threshold for intervention is set at over $1 billion in damages or more than 50 people being killed or seriously injured.

The proposed legislation would cover various risks associated with advanced AI models, including those related to biological weapons, an AI model circumventing its developer’s safeguards, and cyberattacks. Under the framework, developers of frontier AI systems would be required to test their models, disclose their findings, and subject them to independent assessments. This would involve periodic evaluations by independent evaluators registered with the Commerce Department.

Companies developing advanced AI models could pay for these assessments, but only if the compensation is not linked to the outcome of the evaluation. Evaluators who identify an imminent risk meeting the legislation’s specified criteria would have to notify the Commerce Department within 72 hours of completing their assessment. The department could then ask a judge to suspend a model if it determines that the system poses the requisite threat.

The draft bill reportedly limits the authority of other federal departments, agencies, and government officials from independently suspending a model’s development, deployment, or internal use. A federal judge in the US District Court for the District of Columbia could order a suspension of the model. This proposal comes as governments grapple with how to regulate increasingly capable AI systems while balancing security concerns with continued technological development.