Anthropic Proposes New Legislation for AI Regulation

As artificial intelligence technology advances rapidly, safety remains a critical priority. Anthropic, a leading company in the field, has developed a legislative initiative that would empower the government to block the release of AI models capable of causing potentially catastrophic harm.## What are the risks?Under the proposed bill, regulations would be triggered by a "foreseeable and imminent risk." This criterion covers scenarios where the use of a model could result in the deaths of more than 50 people or property damage exceeding $1 billion. The bill specifically highlights threats related to the creation of biological weapons, models bypassing control mechanisms, or the execution of cyberattacks.These figures—50 people and $1 billion—are already utilized in California state legislation (Senate Bill 53), which defines such incidents as "catastrophic risks." Anthropic's initiative seeks to implement this approach on a broader, federal level.## How will models be evaluated?The bill proposes the creation of an "independent evaluator" institution, to be registered with the U.S. Department of Commerce. While AI labs themselves might fund these evaluators, the bill prohibits payment structures that are contingent upon the results of the assessments.According to the procedure, if independent experts identify a catastrophic risk, they would have 72 hours to notify the Department of Commerce. The department would then be required to petition a court to halt the distribution of the model. A key detail is that the bill explicitly forbids federal agencies from unilaterally blocking models, emphasizing the necessity of judicial oversight in the decision-making process.## Why does this matter?The tech industry is currently divided. Some companies, including Nvidia, argue that new laws are unnecessary. Conversely, Anthropic is attempting to regulate the very sector in which it operates through its "Advanced AI Framework."While the threat of "agentic hacking"—where AI programs hack websites without human intervention—has recently become a major concern, Anthropic's proposal focuses on broader, existential risks. This initiative represents another step toward AI developers seeking to establish legal frameworks rather than relying on voluntary monitoring, in an effort to preempt more restrictive, uncontrolled regulations in the future.