Artificial intelligence firm Anthropic has discreetly submitted a draft legislative proposal to United States lawmakers aimed at setting new safety mandates for advanced AI systems. According to an October 16 report by Politico, the draft also incorporates restrictions regarding when the government would be permitted to suspend the release or utilization of these technologies.
As detailed in the report, the framework would restrict government intervention exclusively to scenarios where an AI model presents a foreseeable, imminent threat capable of causing over USD 1 billion in damages, or resulting in the death or severe injury of more than 50 individuals.
AI Safety Bill Sets Thresholds
Politico noted that the suggested limitations would target hazards associated with biological weapons, cyberattacks, or situations where AI models break free from the control of their creators. Furthermore, the legislation would mandate that creators of the most advanced AI systems perform safety evaluations, share their results, and secure independent reviews.
A representative for Anthropic informed Reuters that the draft mirrors protocols detailed in the organization’s Advanced AI Framework, which was published earlier this year. The spokesperson stated that the bill aims to grant authorities the “ability to block or deter dangerous deployments of the most powerful models” while simultaneously enforcing developer testing and independent assessments.
Additionally, the legislation would create a structured mechanism for evaluating prospective dangers. Under this system, independent assessors would register with the US Department of Commerce to carry out routine inspections of AI creators and their cutting-edge models.
Independent Reviews and Court Oversight
Politico reported that AI firms would be allowed to fund the independent assessors, provided that their compensation is not contingent upon the assessment’s conclusions. Any evaluator who detects an “imminent enumerated risk” would be legally obligated to inform the Commerce Department within 72 hours of finishing their examination.
Following notification, the department would have the authority to petition a judge for a model suspension if it concludes the system presents a qualifying hazard. The text would explicitly prohibit other US government agencies, departments, and officials from independently halting AI development, deployment, or internal application. Judicial orders for suspensions would need to come from a federal judge in the District of Columbia, the report stated.
Reuters reached out to the US Department of Commerce for comment, but did not receive an immediate response. While the draft highlights Anthropic’s ideal framework for regulating artificial intelligence, the report did not clarify whether members of Congress intend to embrace the draft or move it forward as official legislation.
Also Read: Anthropic Opens Advanced Claude Models to More Cybersecurity Firms
