The Trump administration is now considering the establishment of its own regulatory authority for particularly powerful AI models. According to a report by the news agency Bloomberg, Treasury Secretary Scott Bessent is working on a proposal modeled after the securities regulator Finra that would be subordinate to the Securities and Exchange Commission (SEC). He is picking up on a proposal by Google DeepMind CEO Demis Hassabis, which beckmann.ai already reported on July 15.
Bessent plan contradicts the government’s earlier line
The new draft marks a shift in course: a departing White House tech advisor had stated in early July that there would be “no FDA for AI,” arguing that a central approval process would mostly create bureaucracy. Now, according to Bloomberg, the government is itself working on an authority that would formally report to the SEC but operate independently – similar to how Finra oversees the securities industry rather than a traditional federal agency handling every review. How this new line squares with the earlier rejection of a central approval body remains unclear.
Bloomberg cites government officials familiar with the process; an official confirmation from the White House is still pending, and the existence of the draft has not been independently verified. Industry is to be involved in financing and setting testing standards, but details on powers, staffing, budget, and the exact relationship with the SEC remain open. A Treasury spokesperson has not publicly commented on the report.
Wiles reviews draft, Trump not yet briefed
White House Chief of Staff Susie Wiles is currently reviewing the plan, and President Trump has not yet seen it, Bloomberg reports, citing several people familiar with the matter. Former White House AI advisor David Sacks is also still advising on the topic, according to Business Standard. Triggers cited include industry criticism of earlier case-by-case reviews at Anthropic and OpenAI, as well as competitive pressure from the Chinese language model Kimi K3, whose unveiling in Washington was seen as an additional warning sign because it briefly suggested a capability edge for Chinese providers.
Hassabis himself had already published his framework, named the “Standards Body,” on July 14: labs would initially submit their models voluntarily 30 days before release for review, with testing covering cyber and bio risks as well as the ability of AI agents to bypass or deceive safeguards. If the process proves effective, Hassabis says, it could become a mandatory rule for the US market. Details on structure, funding, and testing criteria are covered in the earlier post.
Nadella, Altman, and Musk back the proposal
Unlike just a week ago, several major providers have now publicly voiced support. Microsoft CEO Satya Nadella and OpenAI CEO Sam Altman have endorsed Hassabis’ approach, according to TechCrunch, as has Tesla and xAI founder Elon Musk. Anthropic and OpenAI initially did not comment on the proposal.
The broad industry support has likely increased pressure on the government to develop its own position rather than leaving the field solely to Google DeepMind. Bessent’s draft and Hassabis’ model pursue similar but not identical goals: the government plan would create a state-anchored authority, while Hassabis leans more toward industry self-regulation with government backing.
It remains open whether Trump will take up the Bessent draft at all, and how two parallel proposals – one governmental, one industry-led – might eventually be merged. Whether a US authority would have any real leverage over models from China or the EU outside US distribution channels is also unanswered in both drafts.
