An article from Politico published yesterday reconstructs in considerable detail one of the most significant and chaotic moments in the Trump administration’s technology policy: the president’s sudden decision on May 21 to block at the last minute the signing of an AI executive order.
This episode not only derailed a measure that had been prepared for months but also brought to light the deep divisions within the West Wing on how to address a crucial technology.
Based mainly on the testimonies of two senior White House officials who spoke anonymously to discuss delicate internal dynamics, and another source close to the administration, the piece offers a vivid picture of the different factions at play and the reasons that led to a postponement which, according to the same sources, does not amount to a definitive cancellation.
Three schools of thought
Within the administration, there are at least three main positions, strongly at odds with each other.
On one side is the “pro-industry” front, clearly embodied by David Sacks, former White House AI lead. This group advocates a minimalist approach to regulation, believing that any excessive constraints risk slowing down American companies in the global geopolitical challenge, primarily against China.
It was Sacks himself who personally intervened with a phone call to President Trump, derailing the signing of the executive order due to concerns expressed by the tech world, which feared overly burdensome requirements for a sector still in its maturation phase.
On the opposite side are the AI hawks, worried about the risks of malicious use of the technology. Among them stand out Secretary of Defense Pete Hegseth and his undersecretary Emil Michael, a former Silicon Valley executive. This camp pushes for stronger barriers, especially against advanced models like Mythos, fearing they could end up in the hands of strategic rivals like Beijing and be exploited for destructive or espionage purposes.
In an intermediate position are Chief of Staff Susie Wiles and Treasury Secretary Scott Bessent. They support a regulatory framework based on voluntariness: companies should offer the U.S. government a “first look” at new models before their public release, without resorting to formal legal obligations. This compromise line seeks to reconcile innovation and national security protection.
These internal divisions clearly demonstrate how the Trump administration’s AI policy is not the result of a predefined unified vision but is taking shape day by day, responding to continuous developments from the sector.
What the blocked executive order entailed
The executive order Trump was about to sign did not have the contours of heavy or intrusive regulation. It was rather a system of voluntary oversight: AI companies were invited to consult with the government on their most advanced models, allowing federal authorities to review them in advance to identify any risks of exploitation by malicious actors.
As one of the interviewed officials explained, it was not configured as “a government telling companies what they can or cannot do,” but as a request for priority access. The main companies in the sector had already informally agreed to respect this mechanism.
Despite the sudden reversal, this approach, or a slightly modified version of it, remains on the table. Internal sources believe there is still room to convince the president, possibly by making small tweaks to individual clauses.
The triggering factor: the arrival of Mythos
A central element of this story is the role played by Anthropic’s Mythos model.
It is an extremely advanced AI system, capable of identifying and exploiting unknown vulnerabilities in any computer system. Its public launch dramatically accelerated the administration’s timeline.
As early as February, a first draft of an executive order more focused on cybersecurity had circulated: security standards for public models, standardization of secure development processes, funding for defenses against AI-enabled attacks, and greater involvement of the National Security Agency.
Between late February and March, confidential briefings began for Susie Wiles, Scott Bessent, and Vice President JD Vance on the model’s real capabilities and what labs were developing internally.
The release of Mythos generated strong concern among several senior officials, including Wiles, described as more oriented toward a hardline stance than publicly apparent, especially regarding dangers to critical infrastructure and national security.
Industry contacts and confidential meetings
Since late April, the Office of the National Cybersecurity Director, led by Sean Cairncross, has taken charge of relations with major tech companies.
Individual and collective meetings followed with OpenAI, Google, Anthropic, and others, during which companies provided feedback and indicated their non-negotiable points. Executives were asked to sign confidentiality agreements from the start of the meetings.
However, the final version of the executive order remained confidential and did not follow the usual inter-agency process.
On May 20, two joint review sessions were held at the Eisenhower Executive Building: executives were able to examine a paper copy of the document and, at least apparently, all seemed in agreement. The signing was scheduled for the following day.
Sacks’ decisive intervention
The plan fell apart when Sacks called Trump on May 21, a few hours before the signing, convincing him it was a mistake.
According to sources, Sacks had initially supported the initiative but then changed his mind, believing the approach was not light enough compared to the more restrictive positions supported, for example, by the Pentagon.
After the long Memorial Day weekend, the administration returned to square one. All options are now on the table: discussions include introducing a 90-day government review period before the public release of models, or eliminating the voluntary prior consultation part altogether, limiting the executive order to creating a cybersecurity coordination center for AI led by the Treasury to coordinate security updates on vulnerabilities and other operational measures.
Future prospects
White House spokesperson Liz Huston emphasized that the administration is working to maintain a balance between innovation and security, reiterating that “America is winning the global AI race.”
A Treasury spokesperson echoed this message, closely linking economic security and national security.
Trump’s reversal therefore does not seem to mark the end of the game but rather the opening of a new round of internal negotiations.
