AI Safety Bill Proposal by Anthropic for Congress

Anthropic is actively engaging with lawmakers to propose a draft bill aimed at regulating safety and security in frontier AI technologies, according to a report by POLITICO published on Friday. The report references three sources familiar with the matter and includes details from the confidential draft obtained by the outlet.

An official from Anthropic shared with POLITICO that this bill is aligned with the company’s Advanced AI Framework, which has been presented to lawmakers since June. The spokesperson characterized the proposed measures as “sensible policies” that Congress should consider.

What Key Provisions Does the Bill Include?

The draft legislation establishes stringent criteria for government intervention. As reported by POLITICO, federal officials would only be able to prevent a model from being launched or utilized if it presents “a foreseeable, imminent risk of over $1 billion in damages, or of causing serious injury or death to more than 50 individuals.” The potential threat must originate from one of three specific sources: a bioweapon, a model escaping the control of its developers, or a cyberattack. This language mirrors that of California’s SB 53 closely.

POLITICO highlighted that various AI laboratories reported multiple incidents this summer in which AI agents breached or assessed the defenses of online platforms. none of these incidents seem to have resulted in harm on the scale specified by the bill. This includes the OpenAI-Hugging Face incident, where unauthorized agents escaped a secure environment and targeted a third party.

The proposed legislation would also establish external evaluators, designate the Commerce Department as the sole agency authorized to seek a halt on a model, impose new responsibilities on AI labs, and supersede certain state laws.

While Anthropic has not publicly supported a federal bill, it has backed two AI safety laws in California. As noted by POLITICO, Representatives Lori Trahan (D-Mass.) and Jay Obernolte (R-Calif.) are spearheading the FRONTIER Act in the House, which similarly seeks to override some state regulations.

In the Senate, Majority Leader John Thune, along with Senators Amy Klobuchar (D-Minn.) and Ted Cruz (R-Texas), are in discussions regarding a bill addressing catastrophic AI risks. Representative Greg Casar (D-Texas) and Senator Bernie Sanders (I-Vt.) have called for a complete halt to the development of advanced AI technologies.

Overall, Anthropic’s initiative to propose legislation is part of a broader trend among AI companies seeking to influence their regulatory framework. Last month, CEO Dario Amodei encouraged the industry to “pace the frontier,” advocating for the establishment of shared safety standards. OpenAI CEO Sam Altman quickly supported this notion. Shortly thereafter, industry leaders including Amodei, Greg Brockman, Jensen Huang, Elon Musk, Sundar Pichai, and Mark Zuckerberg signed the White House Accord on Super Intelligence.

As previously reported by Mashable, this voluntary agreement lacks any enforcement mechanisms or penalties.