From October 20 to 22 along the banks of Shanghai's Huangpu River, the 2026 Sustainable Global Leaders Conference returns once again. Nobel and Turing Award winners will take the stage to discuss the future and trends of technology, and the internet-famous professor who "rescues" people from anxiety and the era of excessive competition, Macfarlane, will hold a live dialogue with Xiang Biao!
Limited seats available, first come first served. Let us join together for the appointment with sustainability! [Click to register] According to three people familiar with the relevant efforts who were permitted to discuss internal deliberations on condition of anonymity, representatives of AI giant Anthropic have quietly pitched a draft frontier AI safety regulation bill to lawmakers on Capitol Hill. As Congress weighs how to respond to the rapidly developing AI industry, this document represents an important move by this frontier AI lab, an attempt to shape the regulations governing its own technology.
An Anthropic spokesperson said: "This bill reflects the various approaches set out in our Advanced AI Framework, and since June we have been advocating this framework to federal and state lawmakers. We believe these are sound policies, and Congress should take action on them."
Under the bill, the government would only have the authority to block a model's release or deployment if the following conditions are met: the model poses a foreseeable and imminent risk that could cause more than $1 billion in losses, or lead to the death or serious injury of more than 50 people. The risk must stem from the development and release of biological weapons, a model bypassing developer controls, or cyberattacks. The wording of the provisions draws on California AI-related laws, including SB 53, which was signed into law last year.
This summer, several leading AI labs disclosed a series of incidents in which AI agents breached or probed various websites and platforms. None of the cases disclosed so far have caused mass casualties, nor have they produced losses exceeding $1 billion. Anthropic has not publicly stated support for any specific legislative proposal in Congress, but it has backed some state-level AI safety bills, including two in California. The company's head of US state and local government relations previously said he hopes states will continue to introduce stronger AI safety regulations, raising regulatory standards nationwide.
Under this bill, independent evaluation bodies would need to register with the Department of Commerce and would be responsible for regularly assessing AI labs and their frontier models. Evaluation bodies could be paid by the labs, but their compensation must not be tied to the evaluation results. If an evaluation body determines that a company's frontier model poses a "listed imminent risk," it must report to the Department of Commerce within 72 hours after the evaluation is completed. After the Department of Commerce determines that a model poses a risk, it could ask a judge to halt the model's release; however, the text explicitly prohibits any other US "department, agency, or official" from pausing a model's research and development, deployment, or internal use, and only a judge of the US District Court for the District of Columbia would have the authority to issue an order suspending a model.
This summer, Anthropic encountered broad export controls, and two of its most advanced models were forced offline. The relevant decision was finalized after a series of emergency phone calls between government officials and company executives, leaving some in the tech industry worried that a company's access to AI technology could be restricted solely at the personal discretion of government officials.
Anthropic's bill would also preempt some state AI laws in areas involving public disclosure of risks, third-party evaluations, and reporting of major safety incidents. The issue of regulatory preemption is political dynamite on Capitol Hill, with Democrats opposing some proposals out of concern that they would wipe out regulations already enacted in states such as California, Illinois, and New York. A large number of tech industry groups argue that, compared with a single national standard, a patchwork of state regulations would greatly increase compliance difficulty for companies.
The bill also would require frontier AI labs to establish safety frameworks, publish transparency reports for new frontier model releases, set up whistleblower protection mechanisms, and report major safety incidents to the Department of Commerce.
This summer, Capitol Hill saw frequent activity around AI safety legislation. In the House, Massachusetts Democratic Representative Lori Trahan and California Republican Representative Jay Obernolte led the introduction of the FRONTIER Act, which authorizes the government to halt the deployment of high-risk models and would preempt some state AI regulations. In the Senate, Majority Leader John Thune, Minnesota Democratic Senator Amy Klobuchar, and Texas Republican Senator Ted Cruz are negotiating an AI safety bill aimed at reducing catastrophic risks. But the bill has run into resistance: a group of safety-oriented organizations, as well as Maria Cantwell, the top Democrat on the Commerce Committee on which Cruz sits, want to give the government greater authority to review AI models. There are also progressive lawmakers, including Texas Democratic Representative Greg Casar and Vermont Independent Senator Bernie Sanders, whose demands are more radical, calling for a pause on advanced AI research and development and a ban on "superintelligence," a type of artificial intelligence whose cognitive abilities theoretically far exceed those of humans.
Comments