The Trump administration is examining the possibility of introducing government oversight of new AI models before their public dissemination.
As the New York Times reports, which broke the story citing American officials and people familiar with the talks, discussions are underway about a possible executive order to establish a task force bringing together executives from major tech companies and government representatives, aimed at reviewing procedures for new AI systems.
This represents a sharp and significant change from the previous approach of aggressive deregulation that characterized the early months of the second presidential term.
After strongly promoting a hands-off approach, defining AI as a sector to be allowed to grow freely, the White House now seems oriented towards forms of state supervision motivated mainly by concerns related to national security, cyber risks, and potential political repercussions.
Need for rules
Until a few months ago, Trump had presented himself as a great supporter of American AI. Last July he described it as “a beautiful newborn baby” that should be allowed to grow without political obstacles or “stupid rules.”
The president quickly canceled the security measures introduced by the Biden administration and focused entirely on competition with China, loosening environmental constraints and facilitating exports.
Today the landscape appears different. According to the Times, the administration is considering introducing a formal government review process for new models. This turnaround comes as bipartisan concerns grow over risks to employment, privacy, energy consumption, and especially cybersecurity.
The executive order and the task force
At the center of the conversations is a possible executive order to create an “AI task force.” This body would bring together tech company executives and government officials to study oversight procedures.
Last week, senior White House officials met with executives from Anthropic, Google (Alphabet), and OpenAI to outline some of the proposals under consideration, as reported again by the Nyt.
It is not yet clear how the review mechanism would work in detail. One hypothesis is a system similar to the British model, where various government agencies verify that models meet certain safety standards.
The main goal would not be to block publications, but to guarantee the government priority access to assess risks and possible military or intelligence applications.
The Anthropic Mythos case
Much of this new attention stems from Anthropic’s announcement about the Mythos model. This system has proven extraordinarily effective at identifying vulnerabilities in software, so much so that there has been talk of a possible “recognition” in the cybersecurity sector.
The company chose not to release it publicly, limiting its use to a few selected partners.
As Reuters also notes, the White House fears political repercussions in the event of a serious cyberattack facilitated by AI.
At the same time, American officials are evaluating the model’s potential for intelligence agencies and the Pentagon.
The April meeting between Chief of Staff Susie Wiles, Treasury Secretary Scott Bessent, and Anthropic CEO Dario Amodei focused precisely on Mythos.
Tensions with the Pentagon
AI policy is complicated by an open dispute between the Pentagon and Anthropic.
After failed negotiations over a $200 million contract and disagreements about security measures for military uses, the Department of War halted the use of the company’s technologies on classified networks. The situation has created difficulties for government agencies that had become accustomed to these tools.
Meanwhile, the White House’s reference framework has changed. David Sacks, former AI lead and supporter of strong deregulation, left his position in March.
Susie Wiles and Scott Bessent are taking a more active role, trying to mend relations with companies and bring Anthropic technologies back into government use.
Divisions in the tech world
The discussions are creating divisions even among companies. Some executives fear that excessive government control could slow American innovation in the race with Beijing.
Others recognize, however, the need for shared procedures in a sector evolving at a dizzying speed. Dean Ball, former AI advisor to the Trump administration, spoke of a delicate balance between speed and caution.
Speculations?
For now, the White House is keeping a low profile. An official called the rumors about a possible executive order “speculations,” reiterating that any announcement will come directly from Trump.
It remains to be seen whether this phase will lead to a light verification system or something more structured, possibly involving the National Security Agency, the office of the National Cyber Director, or the Center for AI Standards and Innovation.
