Representatives from the artificial intelligence firm Anthropic have reportedly presented a confidential draft bill to members of Congress. According to three anonymous sources familiar with the discussions, the legislation aims to establish new rules for regulating frontier AI safety and security, marking a significant effort by the developer to influence the legal framework governing its technology.
Proposed Regulations and Scope
The draft proposal is designed to regulate how advanced AI models are released and to preempt certain state-level AI safety laws. According to the text, the government would only be permitted to restrict a model’s deployment or usage if it posed a foreseeable and imminent risk meeting specific thresholds. These thresholds require that the risk involves either more than $1 billion in potential damages, or the threat of seriously injuring or killing more than 50 people.
Furthermore, the bill specifies that such risks must originate from three defined sources: the creation or release of a biological weapon; the model successfully evading the developer’s installed controls; or a sophisticated cyberattack. The language used in the proposal closely mirrors existing state legislation, specifically referencing the provisions found in California’s AI laws, including SB 53, which was enacted last year.
Developer Commentary and Advocacy
The confidential draft represents proposals detailed within Anthropic’s Advanced AI Framework. An Anthropic spokesperson confirmed the company’s involvement, stating that the legislation reflects policies the company has been actively advocating for with both federal and state lawmakers since June.
This bill represents the proposals outlined in Anthropic’s Advanced AI Framework, which we have been advocating for with federal and state lawmakers since June. “These are sensible policies we believe Congress should act on.”
