Trump Signs Streamlined AI Executive Order Away from Cameras

On June 2, Donald Trump quietly signed a decree on artificial intelligence, after initially abandoning it two weeks earlier. This now streamlined text marks a step back from the stricter version previously considered.

In January 2025, the Trump administration quickly rescinded Joe Biden's decree on AI security. Sixteen months later, a new text was drafted, but it was abandoned on May 21 under pressure from tech leaders. The decree signed in June, described by the New York Times and the Washington Post, is a softened version of the original.

A Revised and Relaxed Decree

Titled "Promoting Advanced Artificial Intelligence Innovation and Security," the decree retains the initial structure but modifies a key element: the submission of models for government evaluation is now voluntary, contrary to the mandatory requirement proposed in the previous draft.

The text also establishes a cybersecurity hub under the direction of the Treasury Department, aimed at centralizing information on vulnerabilities. Federal agencies are encouraged to develop criteria for assessing the cyber capabilities of models. The signing followed a meeting at the White House with key figures such as Treasury Secretary Scott Bessent, Defense Secretary Pete Hegseth, and David Sacks, who oversees AI and cryptocurrencies in the administration. Initially opposed to the draft, Sacks supported the revised version after the review period was shortened.

Divergent Approaches Between Washington and Brussels

Biden's decree from October 2023 required developers to share their security tests before any risky deployment. Trump, by rescinding this decree upon taking office, opted for a deregulation policy. The return of an evaluation text, even if optional, in a context of growing concerns about the cyber capabilities of models, is noteworthy.

The initially more stringent draft was abandoned due to concerns from tech leaders about American competitiveness against China. The final version, devoid of any obligation, contrasts with the European approach. The EU's AI regulation, set for August 2, 2026, imposes strict obligations for high-risk models, including documentation and penalties. Two philosophies are at odds: optional evaluation in the United States and a binding framework in Europe.