On October 17th, local time, Politico reported, citing three sources familiar with the matter, that representatives from Anthropic had quietly introduced a draft bill to some congressmen, which is said to aim at regulating the security measures for advanced artificial intelligence.
According to the draft proposed by Politico, the government can only prevent the release or use of an AI model if there is a so-called ‘foreseeable immediate risk’, which means that the model could cause losses exceeding $1 billion, or result in the death or serious injury of more than 50 people.
Additionally, the aforementioned significant risks must arise from the creation or dissemination of biological weapons using AI models, from being out of the developer’s control, or from launching cyber attacks.
The draft also stipulates that if an independent assessment agency determines that a company’s advanced model poses the aforementioned risks, it must report to the U.S. Department of Commerce within 72 hours after completing the assessment. The department will then apply to a federal court in Washington, D.C., for a restraining order to suspend the release or use of the model.
However, the draft text clearly prohibits “any other U.S. government departments, agencies, or officials” from arbitrarily suspending or interfering with the development, deployment, or internal use of these models. Politico noted that this summer, Anthropic was affected by extensive government export controls, resulting in the shutdown of its two most advanced models. At the beginning of October, the Pentagon officially ceased using Anthropic’s products.
A spokesperson for Anthropic confirmed that this draft bill reflects the recommendations contained in the company's earlier ‘Advanced AI Framework’. ‘Since June, we have been advocating these recommendations to federal and state legislators,’ said the spokesperson.
He added, "We believe these are reasonable policies and Congress should act."
The Anthropics declined to disclose the scope of this draft and its current acceptance level. The U.S. Department of Commerce also has not responded to requests for media comment.
The technology and policy analysis websites ExplainX and Tech Insider have pointed out that independent researchers in the field of AI safety have publicly criticized the high thresholds set by Anthropic. Critics have joked that, according to these standards, none of the recorded severe safety incidents caused by AI between 2024 and 2026 would meet these thresholds. This means that governments will be limited in most cases.
"A company whose products will be directly regulated by the regulatory framework, is actually the law drafter. This is unheard of in Washington, D.C., according to Tech Insider. Anthropic appears to believe that since federal legislation is imminent, it would be better for them to draft the initial drafts than have to respond to others’ proposals."
