Anthropic Proposes AI Safety Bill With 50-Person Death Threshold
Casablanca – Anthropic has proposed a US federal law that would let authorities ask a court to stop the release of an artificial intelligence model if it poses an immediate threat of killing or seriously injuring more than 50 people, or causing more than $1 billion in property damage.
According to Politico, the draft law, which the AI company has presented to US lawmakers, explains when the government could step in during the development and release of advanced AI systems. TechCrunch reported on October 20, citing an earlier report by Politico.
The proposal allows government action only when risks involve creating or releasing a biological weapon, an AI model escaping its developer's controls, or a cyberattack. The danger must meet the law's standard for a foreseeable and immediate risk of a major disaster.
Under the draft law, independent experts would review AI companies and their most advanced models. These experts would register with the US Department of Commerce and regularly check for possible safety risks.
AI companies could pay for these reviews, as long as the payments do not depend on the results. If experts find an immediate risk covered by the law, they would have 72 hours after finishing their review to report it to the Commerce Department.
The department would then have to ask a federal judge in Washington, DC, to stop the model from being released or used. The draft clearly states that other federal agencies and government officials cannot independently stop a model's development, release, or internal use.
Anthropic Warns AI Could Soon Start Building Its Own Successors
A spokesperson told Politico that the draft turns ideas from the company's Advanced AI Framework into a proposed law. Anthropic has been sharing this framework with federal and state lawmakers since June.
The proposed limits are similar to rules in California's Senate Bill 53, which lawmakers recently passed into law. That law also deals with major risks linked to advanced AI systems.
The proposal comes as technology companies face growing questions about AI safety. Recent incidents involving AI programs that reportedly ignored their operators and tried to hack websites have raised concerns about whether developers have put strong enough safety measures in place.
Anthropic's plan also reflects a wider debate about how much power governments should have over AI development. Technology companies have often preferred voluntary safety measures and independent reviews instead of rules that require government oversight.
Other industry leaders have taken a different view of regulation. Nvidia CEO Jensen Huang recently argued that the AI industry did not need new laws, according to TechCrunch.
Anthropic's draft would create a formal legal process for limiting dangerous AI releases. At the same time, it would set clear limits on when and how the federal government could step in.