Draft Legislation Establishes High Thresholds for Model Bans

Representatives from artificial intelligence research company Anthropic have been actively circulating a draft legislative proposal among Capitol Hill lawmakers. The measure outlines a regulatory structure focused on safety standards and security controls for advanced, frontier artificial intelligence systems.

Under the terms of the proposal, federal regulators would possess restricted power to prevent the launch or operational deployment of an artificial intelligence model. Intervention would be permissible solely if an impending, predictable hazard threatens over $1 billion in total damages, or creates a direct danger of causing serious bodily harm or death to more than 50 individuals.

Categorized Danger Triggers and Testing Mandates

The draft specifies that statutory risks must arise from particular severe scenarios, including:

  • The design, production, or distribution of biological weaponry.
  • Advanced models escaping or actively bypassing the administrative safeguards established by creators.
  • Critical infrastructure disruptions or damages resulting from cyberattack incidents.

The proposed legislative framework incorporates principles previously detailed in the developer's public safety blueprint published earlier this year. Along with establishing parameters for federal intervention, the statutory language mandates that artificial intelligence developers perform rigorous evaluations on frontier systems, maintain operational transparency regarding their assessments, and submit systems to third-party technical reviews.

Regulatory Oversight and Judicial Procedures

The operational framework establishes a system of independent auditors required to register formally with the United States Department of Commerce. These accredited external assessors would perform recurring inspections of artificial intelligence laboratories and their most advanced models. Technology developers would be authorized to finance these inspection services directly, provided that contractual compensation remains entirely divorced from the conclusions of the assessment.

If certified evaluators identify what the draft defines as imminent enumerated risks, they must submit formal notification to the Commerce Department within a strict 72-hour window following the end of their audit.

Government authority to pause development remains heavily concentrated. While the Department of Commerce would hold legal standing to petition for an injunction, the statutory text expressly prohibits any other United States government agency, department, or administrative official from halting internal research, public distribution, or operational usage. Formal orders to suspend an artificial intelligence system would rest exclusively with a federal judge sitting on the United States District Court for the District of Columbia.