Amid growing unease about artificial intelligence, Anthropic has drafted a legislative proposal aimed at curbing the distribution of models deemed to present extreme danger. The company shared the draft with lawmakers, seeking congressional consideration, according to reporting by Politico. The initiative reflects Anthropic's ongoing engagement with policymakers since June, as outlined in its Advanced AI Framework.
The proposal defines a scenario where a model poses a foreseeable, immediate threat that could result in the death of more than fifty individuals or generate property loss exceeding one billion dollars. Such threats must arise from the creation or deployment of a biological weapon, the model escaping its developer's safeguards, or a cyberattack, mirroring thresholds used in recent California legislation known as Senate Bill 53.
To assess these risks, the bill calls for the use of independent evaluators registered with the U.S. Commerce Department. While the evaluators could receive compensation from AI firms, the legislation stipulates that payment must not depend on the outcome of the assessment. Critics note that the draft does not detail mechanisms to prevent financial influence over the evaluations.
If an evaluator determines that a model meets the catastrophic criteria, the draft requires notification of the Commerce Department within seventy-two hours. The department would then be obligated to petition a Washington judge for an injunction to stop the model's release. The bill explicitly bars any federal agency from unilaterally halting distribution, shifting final authority to the judiciary.
Anthropic confirmed the existence of the draft to Politico, describing it as a formal expression of the proposals contained in its Advanced AI Framework. The company did not provide additional comment to TechCrunch. The draft represents a more prescriptive approach than voluntary industry measures, aiming to embed external oversight into the development pipeline.
Industry reactions vary, with some firms favoring voluntary embedded monitors rather than mandatory oversight. Nvidia's chief executive recently argued that new regulations are unnecessary, while other companies have suggested self-governing bodies similar to financial industry regulators. Anthropic's legislative push therefore marks a notable departure from the broader sector's preference for internal, non-binding safety mechanisms.