Trump halts AI executive order, regulation defeated by competitive anxiety

Title: Donald Trump Abruptly Postpones AI Order After White House Infighting

Author: Joe Miller

Source:

Reprint: Mars Finance

Editor’s Note: The Trump administration had originally sought to establish a pre-release safety testing mechanism for cutting-edge AI models, but just a few hours before signing, the executive order was suddenly put on hold.

According to the original plan, leading AI companies such as OpenAI, Google, Anthropic, and xAI would voluntarily share their models with the U.S. government 90 days before the advanced models were publicly released, so they could undergo national security and cybersecurity risk assessments. Trump ultimately refused to approve it, saying he “did not want anything to hinder America’s leadership position in AI.”

Behind this sudden turn is the core contradiction facing U.S. AI policy: the capabilities of frontier models have begun to touch on public risks such as cybersecurity vulnerabilities, loopholes in financial systems, job displacement, and data center expansion. But amid the backdrop of U.S.-China AI competition being viewed as a national strategy, any regulatory arrangement could be interpreted by industry as slowing innovation and weakening America’s competitiveness.

More subtly, the executive order was not originally a mandatory approval system; instead, it was a “collaborative, voluntary” model evaluation framework. In other words, the White House was not trying to directly control model releases, but rather hoped to add a government safety testing step before the models were opened to the public. Even with this relatively mild approach, however, it was still temporarily shelved amid the tug-of-war between safety governance and technological leadership.

AI is moving from a purely technology-industry issue into the intersection of national security, the macroeconomy, and political governance. The controversy surrounding U.S. AI policy is also shifting from “whether to support AI development” to “how to rein in the systemic risks that frontier models may bring without sacrificing leading advantages.”

Below is the original text:

The White House unexpectedly postponed the signing of a long-anticipated executive order on artificial intelligence. Previously, Donald Trump said he “did not like” certain parts of the plan—especially the arrangement for the U.S. government to conduct national security and cyber risk reviews of AI models.

The executive order was originally scheduled to be signed Thursday afternoon. Under the plan, leading AI companies such as OpenAI, Google, and Anthropic would voluntarily commit to submitting their models to the government for review.

Trump’s sudden reversal came after weeks of ongoing disputes within the U.S. government about the scope and boundaries of regulatory scrutiny.

Speaking about the executive order, Trump said: “I don’t like some aspects of it. We’re ahead of China—and ahead of everyone else. I don’t want anything to hinder our lead.” He also said that artificial intelligence is “also bringing a lot of jobs.”

Before the signing ceremony was abruptly postponed, several technology company CEOs had planned to travel to Washington to attend the event with Trump.

The delay comes as multiple polls continue to show U.S. voters are concerned about the impact of artificial intelligence. Many of them support tighter regulation and safety guardrails for this emerging technology.

Public concerns about the security implications of opening up powerful AI models are rising. At the same time, debates over AI’s impact on jobs and the controversy surrounding large-scale data center construction have made the White House’s political calculations even more complicated. Not long ago, the current administration’s stance toward the AI industry was still clearly more positive.

Some of Trump’s allies had called for leading AI models to be brought under U.S. government control. However, other figures in the MAGA camp warned that any measures to restrict AI growth could hurt the U.S. economy.

A recent poll commissioned by the Institute for Family Studies found that 82% of Americans support the White House conducting safety testing on advanced AI models.

The executive order was developed after key White House officials previously gained early access to Anthropic’s latest Mythos model, which included U.S. Treasury Secretary Scott Bessent. The model has advanced capabilities and can be used to identify cybersecurity vulnerabilities. Officials familiar with the matter said they were shocked by issues such as banking system vulnerabilities exposed by the model.

So far, Anthropic has limited Mythos access to a small number of trusted institutions, including technology companies and some banks, so those organizations can identify and fix cybersecurity problems before hackers gain access to the model.

Kevin Hasset, director of the White House National Economic Council, had once proposed that frontier AI models undergo a process similar to drug approval—only after they are “proven safe, like FDA approval,” could they be officially released.

His statement faced strong opposition from AI founders and investors, including some people closely connected to the Trump administration. They argued that such a system would weaken America’s innovation capacity.

The executive order falls far short of the requirements of a process like formal approvals. Instead, it shifts to establishing a “collaborative, voluntary” model benchmarking framework. On Thursday morning, before the scheduled signing ceremony, White House officials provided reporters with a brief overview of the order’s contents.

Under the proposed agreement, leading AI companies—including OpenAI and xAI—will voluntarily share their models with the government 90 days before the relevant models are released to the public. In other words, this mechanism ultimately still depends largely on the willingness of AI company executives to cooperate.