[DigitalToday reporter Choi Jae-won] U.S. AI company Anthropic has proposed to Congress a draft bill that would allow the government to halt the release of AI models that could cause large-scale loss of life or economic damage. The draft strictly limits the conditions for government intervention, raising expectations of controversy over the effectiveness of regulation.

Business Insider reported on Oct. 19 local time that Anthropic recently provided U.S. congressional staff with a draft federal bill to regulate the safety and security of cutting-edge AI models.

At the core are the criteria for the government to block the release or use of an AI model. Intervention would be allowed only when there is a foreseeable and imminent risk that the model could cause more than 50 deaths or serious injuries, or property damage of more than $1 billion. The draft also limits the types of risk to biological weapon development and dissemination, AI operation beyond a developer's control, and cyberattacks.

The draft also introduces an independent safety evaluation system. External evaluators registered with the U.S. Commerce Department would regularly inspect AI developers and models. If they find a serious risk, they must report it to the Commerce Department within 72 hours of completing the evaluation. Evaluators may be paid by AI companies, but the amount must not vary based on the evaluation outcome.

If the Commerce Department confirms a risk, it can ask the federal district court in Washington, D.C. to stop a model launch. The draft restricts other federal agencies from independently halting development, distribution or internal use.

The draft also includes requirements for AI companies to build safety management systems and disclose transparency reports, protect whistleblowers, and report serious incidents.

A key issue is the relationship between federal regulation and state law. The draft would apply federal law first in some areas, including risk disclosure, third-party evaluations and incident reporting. That has raised concerns that state governments' regulatory authority could weaken. Anthropic said it is a reasonable proposal reflecting the AI policy framework it disclosed in June.

The draft is not a formally introduced or enacted law. Negotiations are continuing in the U.S. Congress over AI safety regulation and the scope of government intervention.