Trump Postpones AI Executive Order Hours Before Signing: 90-Day Review for Models Divided White House and Industry
The US President cancelled on Thursday the signing ceremony of an executive order that would create a federal safety review mechanism for AI models before their launch, citing risks of slowing down America's advantage over China.
On the afternoon of Thursday, May 21, US President Donald Trump cancelled the signing ceremony for an executive order on artificial intelligence scheduled for the same day. Invitations had already been sent to executives from the industry. In justifying the postponement to the press, Trump stated that he made the decision because he "did not like certain aspects" of the text and did not want to "do anything that could hinder" America’s advantage over China in AI.
The executive order proposed the creation of a federal safety review mechanism for large-scale AI models prior to their public launch. The circulating version established a review period of up to 90 days during which federal agencies could evaluate models before their commercial availability. OpenAI and Anthropic, the two companies most involved in negotiations with the White House, advocated for significantly shorter review periods.
What the Executive Order Proposed
The proposal would create a voluntary "clearinghouse" coordinated by the Department of the Treasury along with other federal agencies. Through this mechanism, developers of advanced models would share pre-launch versions with the government for analysis of vulnerabilities and security risks. The stated goal was to protect government networks from AI-enabled attacks and allow the federal government to map the capabilities of models before their public dissemination.
The central impasse was the timeline. A 90-day review period would make it unfeasible for the cadence of releases that leading AI companies need to maintain to compete globally. Anthropic, whose annualised revenue surpassed $2 billion in early 2026 and relies on rapid release cycles to renew contracts with partners such as Google and Amazon, was among the most active in negotiations to shorten the proposed timeline.
The Logic Behind the Postponement
Trump framed the decision in geopolitical terms: "The United States is ahead of China and the rest of the world in AI." The statement signals that, in the short term, the White House will prioritise speed of innovation over formal pre-launch oversight structures. No schedule for rescheduling the signing was announced. Administration aides only mentioned that the text remains "under review."
Industry pressure on the White House was direct: companies argued that a 90-day period per model would create an unprecedented regulatory bottleneck in the history of American software, at a time when European and Asian competitors do not face equivalent pre-launch restrictions. The presidency did not disclose which specific aspects of the text prompted the last-minute decision.
Implications for CISOs and Consulting Firms
For corporate security teams, the absence of a federal pre-review standard for models places the onus of due diligence on buyers of AI technology. Without uniform requirements for disclosing capabilities or vulnerabilities, companies integrating large language models into critical systems must conduct proprietary assessments, increasing the cost and heterogeneity of implementations.
For consulting firms with federal contracts, the suspension paves the way for security requirements defined on a contract-by-contract basis, increasing compliance complexity in government projects involving AI models. Without a federal standard, agencies such as the Department of Defense and the DHS will tend to impose individual security clauses in each contract, fragmenting the governmental AI solutions market.
In the same week, Congress made no advance on any of the pending federal AI legislation proposals since 2025. With midterm elections in November 2026, any federal regulation of AI will primarily depend on executive action, making Trump’s position on the issue the main regulatory variable for the sector in the next twelve months.
The contrast with the European Union is immediate: the AI Act made rules for general-use AI models mandatory in August 2025, and the European Commission is maintaining an open public consultation on high-risk systems classification guidelines until June 23, 2026. Companies developing models for both markets will need to navigate between a bloc that has not created formal pre-launch barriers and another that requires detailed technical documentation before deployment, an asymmetry that favours the location of R&D centres in the US while the European regulatory debate progresses.