Anthropic has submitted a draft bill to US lawmakers aimed at regulating AI safety. The proposal suggests that federal judges should hold the exclusive authority to block the release of advanced AI models, provided that the potential damage exceeds $1 billion or results in at least 50 deaths or serious injuries.
The draft outlines specific risks, including the development of biological weapons, loss of control over model distribution, and the execution of cyberattacks. Anthropic proposes that independent auditors, registered with the US Department of Commerce, conduct periodic checks on AI developers. The company has offered to fund these audits itself.
Under the proposed framework, auditors must report imminent risks to the Department of Commerce within 72 hours. The department would then have the right to seek a court injunction, but only a federal judge in the District of Columbia would be authorized to issue a final ban on further development.
This initiative follows Anthropic's previous lobbying efforts at the state level in California. The company aims to establish unified federal regulations to replace the current fragmented landscape of state-level policies, while also proposing protections for whistleblowers and standardized reporting requirements for critical incidents.
