President Donald Trump pulled the plug on an artificial intelligence executive order just hours before he was set to sign it in late June 2026, telling reporters that a mandatory pre-release testing requirement for AI models could “block” the United States from maintaining its lead over China. The decision, which caught parts of the White House staff off guard, effectively shelves a draft order that would have required leading AI developers to submit their most powerful systems for government review before public release.
The reversal sharpens a tension that has defined the administration’s approach to technology: how to balance national security oversight with the breakneck speed American companies say they need to stay ahead of Beijing.
What the shelved order would have done
The draft executive order contained what reporting from the Associated Press and the Washington Post described as an up-to-90-day pre-release review window. Under that provision, federal agencies would have tested frontier AI models before companies could ship them to the public. Trump reportedly described that review window as a “blocker” and told reporters he did not want to “get in the way” of American AI progress relative to China.
The full text of the draft has not been released, so key details remain unclear: whether the review would have applied to all frontier models or only those above a specific capability threshold, whether the testing was strictly mandatory or could function as a voluntary preview, and which agencies would have conducted the evaluations. No named officials have gone on the record describing the internal debate over those design choices.
No official White House transcript of Trump’s remarks has been published. The quotes attributed to him come from press pool reports rather than a formal record, leaving some ambiguity about whether he indicated a timeline for revisiting the order.
How it fits the administration’s AI trajectory
The postponement did not arrive in a vacuum. It sits at the end of a policy arc the administration has been building since Trump took office.
In January 2025, the White House released a fact sheet on AI leadership that framed the administration’s priorities around reducing regulatory burdens on developers and accelerating private-sector innovation. That same month, Trump revoked the Biden-era Executive Order 14110, titled “Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence,” which had been issued in October 2023. Biden’s order had imposed reporting obligations, model evaluations, and standards development for AI systems deemed to pose national security or safety risks.
By mid-2025, the administration went further with what it called America’s AI Action Plan, which explicitly organized federal policy around “winning the AI race” against China. The now-shelved executive order was supposed to be the next step in that sequence, layering government testing on top of the competitive framing. Instead, Trump decided the testing layer contradicted the speed-first philosophy he had endorsed for more than a year.
What major AI companies have not said
Notably absent from the public record is any formal reaction from the companies that would have been most directly affected. OpenAI, Google DeepMind, Anthropic, and Meta all operate frontier AI programs that likely would have fallen under the order’s scope. None have issued public statements about the postponement.
That silence is itself significant. In July 2023, several of these same companies made voluntary commitments at the White House to conduct pre-release safety testing on their most powerful models. Those pledges were made under the Biden administration, and it remains unclear whether the companies still consider them binding or whether the commitments have quietly lapsed alongside the regulatory framework that accompanied them.
The timing of the shelved order has also prompted speculation about commercial pressures. Several leading U.S. AI labs are expected to release new frontier models in the coming months, and a 90-day federal review window would have introduced real friction into those product timelines. No administration official has confirmed on the record that release schedules influenced the decision.
The gap between ambition and infrastructure
Even if Trump had signed the order, serious questions would have remained about whether the federal government was prepared to carry it out. A meaningful pre-release testing regime would require technical benchmarks, red-teaming protocols, and clear criteria for what constitutes an unacceptable risk. None of that groundwork has surfaced publicly.
The agencies most likely to have been tapped for the work, such as the National Institute of Standards and Technology and the Department of Homeland Security, both played roles in Biden-era AI safety efforts. Neither has commented publicly on the shelved order or described any preparations for taking on a new testing mandate. No implementation memo or budget request has appeared, leaving outside observers with little sense of whether the 90-day window was envisioned as a heavily resourced program or a lighter procedural check.
By contrast, the European Union’s AI Act, which began phased implementation in 2024, includes pre-market conformity assessments for high-risk AI systems and has dedicated institutional resources to enforcement. The gap between that structured approach and the ad hoc nature of the shelved U.S. order underscores how far behind Washington remains in building durable AI governance infrastructure, even as it leads in AI development.
What developers and the public should watch for next
For AI developers, the immediate signal is clear: the administration remains reluctant to impose pre-release constraints on AI deployment, even when national security officials raise concerns about misuse. Companies can continue to structure their roadmaps around internal safety testing and market-driven pressures rather than a formal federal review calendar. That may encourage faster iteration, but it leaves questions about systemic risk management to voluntary industry coordination and narrower sectoral rules.
On Capitol Hill, the shelved order may give new momentum to lawmakers who have argued that AI oversight should come from Congress rather than executive action. The Senate’s bipartisan AI working group has circulated multiple policy frameworks, though none has advanced to a floor vote. The spectacle of a president drafting and then abandoning his own safety measure could strengthen the case that legislation, not executive orders, is the only way to create rules that survive a single news cycle.
The deeper takeaway is how unsettled U.S. AI governance remains. The same White House that made “winning the AI race” a central theme briefly entertained a significant new oversight tool, only to abandon it at the last moment over fears it might slow domestic innovators. The episode reveals that some advisers are willing to contemplate stronger guardrails for frontier AI, while Trump himself remains focused on avoiding anything that could be cast as a brake on American technological momentum.
Until that tension is resolved in something more durable than a draft order, the rules governing the most powerful AI systems will stay fragmented. And the next ambitious safety proposal could prove just as vulnerable to an eleventh-hour reversal.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
