Donald Trump’s June 2 executive order on artificial intelligence is not a plan. It is a course correction without an apology — a quiet admission that the man who repealed Biden’s AI safeguards on his first day in office spent eighteen months getting it wrong. And the people who pay for that lost time are not in the Oval Office.
On Tuesday afternoon, in a private signing the White House did not broadcast in real time, President Donald Trump put his name to an executive order titled Promoting Advanced Artificial Intelligence Innovation and Security. The order asks — does not require, asks — that America’s frontier AI companies hand the federal government a preview of their most powerful systems up to thirty days before the public ever sees them. According to CNBC’s reporting on the signing, the framework is built around a benchmarking process led by the National Security Agency to flag a model’s “advanced cyber capabilities” before release. The administration is calling it a landmark. It is, in fact, a confession.
Because eighteen months ago, on January 20, 2025, this same president walked into the same building and rescinded Joe Biden’s 2023 executive order on artificial intelligence — an order that required AI developers to share safety-test results with the federal government before unleashing powerful systems on the public. Trump’s allies, as TechCrunch documented at the time, called Biden’s reporting requirements “onerous.” They were thrown into the regulatory bonfire alongside dozens of other Biden directives. Eighteen months of compounding national security risk later, the same provisions are back — voluntary, declawed, and dressed up as innovation.
This is what passes for governance in 2026: a presidency that tears down the safeguards on Monday and rebuilds them, smaller, on Tuesday — and then asks Americans to applaud the construction.
1. The Flip-Flop, On Display
The chronology alone tells the story. The original signing ceremony for this AI order was scheduled for May 21, 2026. As CNN reported, tech executives were already en route to the White House when Trump abruptly pulled the plug, telling reporters he “didn’t like certain aspects” of the order and worried it would “get in the way” of American AI development. Twelve days later, he signed a version that had been shortened, softened, and stripped of teeth — the review window cut from ninety days to thirty after AI czar David Sacks, Elon Musk, and Meta CEO Mark Zuckerberg reportedly called the White House to lobby against the longer timeline.
“Once again, the Trump administration has belatedly discovered the need to redo something it hastily dismantled in its first year. While this course correction can begin to grapple with widespread impacts that new frontier models will have on our critical infrastructure, it can’t undo the years wasted on dismantling some of the most vital pillars of our nation’s cybersecurity response.”
— Sen. Mark Warner (D-VA), Vice Chairman, Senate Intelligence Committee
Warner’s statement, issued the same day the order was signed and posted to his Senate page, is not partisan venting. It is a forensic account of what happened. The Vice Chairman of the Senate Intelligence Committee — a Democrat, yes, but one who has spent years working bipartisan AI legislation — is saying out loud what the technology pages keep euphemizing: the President of the United States made a national-security decision on Day One based on a campaign-trail grievance against Biden, and the country has been catching up to his error ever since.
2. What the Order Actually Does — and Doesn’t
Strip away the rhetoric and the order is, in substance, modest. As NPR summarized, it directs federal agencies to develop benchmarks for “advanced cyber capabilities,” asks frontier labs like Anthropic, OpenAI, and Google to voluntarily share unreleased models with the government for up to thirty days, and creates an “AI cybersecurity clearinghouse” to centralize vulnerability information. The order explicitly forbids any mandatory licensing or preclearance regime. The White House said on social media, in apparent anticipation of right-wing backlash, that “we are NOT conducting oversight of all new models.”
What the order does not do is at least as instructive as what it does. It does not require companies to participate. It does not set criteria for what counts as a “covered frontier model” — that work is punted to the agencies for sixty days. It does not address algorithmic bias, labor displacement, civil rights, deepfakes, or election integrity, all of which were front-and-center in the Biden order Trump tore up. As CyberScoop noted, the document spends its opening section reassuring industry that “we refuse to stifle this innovation with overly burdensome regulation.”
What Changed His Mind
The administration’s pivot from “no oversight” to “voluntary oversight” was driven by the April debut of Anthropic’s Claude Mythos Preview, a model so capable at finding software vulnerabilities that it spooked both Wall Street and the national-security establishment. NBC News reported that Mythos “upended the White House’s calculus.”
Who Was in the Room
According to CNBC, AI czar David Sacks, Elon Musk, and Mark Zuckerberg personally lobbied to gut the May draft. Sacks had argued throughout 2025 that state and federal AI rules were a “threat to America’s AI industry.” The thirty-day window is their handiwork.
The Voluntary Loophole
The order is built on “voluntary collaboration.” As Juan Londoño of the libertarian-leaning Cato Institute told PBS NewsHour, the discretion given to the NSA director is a “dangerous precedent” that could “weaponize” the policy against companies the administration is feuding with.
The Anthropic Question
The same Pentagon that wants Anthropic’s Mythos for testing blacklisted Anthropic last year as a “supply chain risk” over a disagreement about military-use guardrails. Samir Jain of the Center for Democracy & Technology warned the order must not become a tool for the administration to punish companies “for political or other arbitrary reasons.”
3. The Risk Nobody at the Signing Mentioned
Here is the question that should have dominated Tuesday’s press briefing and did not: what happens to the most valuable trade secrets in the American economy once they are inside a federal building?
Frontier AI models are not paperwork. They are model weights — terabyte-scale parameter files that, in the hands of a hostile actor, can be repurposed in days. Stripping the safety guardrails off a stolen model, according to AI-policy researchers, can cost as little as two hundred dollars. The risk is not theoretical. In July 2025, a staffer at the Trump-Musk Department of Government Efficiency accidentally leaked access to at least 52 of xAI’s internal large language models. In 2023, a hacker breached OpenAI’s internal communications and exfiltrated model-design details. Researchers from the UK’s Royal United Services Institute warned in May, in a report covered by The Register, that frontier AI safety tests are “creating the very risks they’re meant to stop” — that the act of giving outside reviewers access to unreleased models multiplies attack surface in ways the industry has not mapped.
And now the United States government — an institution whose own Office of Personnel Management was breached by Chinese intelligence, whose Treasury was hacked in 2024, whose Department of Homeland Security has been hollowed out by mass firings under this administration — is asking to hold the keys to America’s most consequential intellectual property. Not because the administration has built the security infrastructure to handle it. But because the President, after eighteen months of insisting no such infrastructure was needed, suddenly decided otherwise.
“Effective safety frameworks should continue to be developed through democratic institutions, informed by technical expertise and broad stakeholder input, to promote accountability and public trust.”
— Chris Lehane, Chief Global Affairs Officer, OpenAI
Lehane’s statement, delivered the day of the signing, is corporate-careful — but its meaning is clear. The most powerful AI company in the country is publicly asking the executive branch to remember that “democratic institutions” exist. That is what it has come to.
4. What It Means for the Average American
For ordinary people, the question is not whether the government should be looking at AI systems before they ship. Of course it should. Biden’s order said so in 2023. Senate intelligence bills said so for three years running. The question is whether this government, run by this president, is competent to do the looking.
An AI system trained to find security vulnerabilities is, by its nature, a dual-use weapon. In good hands, it patches the power grid. In bad hands, it takes the grid down. The thirty-day pre-release window is the difference between defenders and attackers having the same intelligence at the same moment — and a Trump administration that fires inspectors general for sport, blacklists companies over personal grudges, and stocks the NSA leadership with loyalty hires is precisely the wrong custodian for that intelligence. The cost of getting it wrong will not be borne by David Sacks or Elon Musk. It will be borne by the hospital whose records are ransomed, the small business whose payroll is drained, the senior citizen whose Social Security is rerouted, the voter whose ballot is sown with doubt.
And there is a deeper civic cost. The order does nothing — nothing — about the AI harms that touch most Americans most often: hiring algorithms that quietly screen out Black applicants, lending models that redline by zip code, deepfakes weaponized against women and political dissidents, surveillance tools deployed against immigrants and protesters. Biden’s order at least gestured at those problems. Trump’s order treats them as if they do not exist, because in his political coalition, they do not.
Get Involved Today
Contribute to our mission and turn your concerns into action.
5. What This Says About Leadership
The pattern is the part that matters. In December 2025, Trump signed a separate executive order preempting state regulation of AI — a giveaway to industry so brazen that Senator Edward Markey (D-MA), in a statement from the Commerce Committee, called it the “return on investment” for the president’s billionaire benefactors, delivered “at the expense of our kids, our communities, our workers, and our planet.” Six months later, the same president signs a different order pulling federal oversight back in — but only the cybersecurity slice his donors agreed to. The contradiction is not policy. The contradiction is the policy.
Leadership is not a press release. Leadership is the discipline to act on a coherent set of principles, communicate them honestly, and accept the political costs. What the country has instead is a president whose AI position changes with whichever billionaire spoke to him last, whose signing ceremony for a national-security order had to be canceled hours before it began because he had not read what he was about to sign, and who treats the federal government’s relationship to the most transformative technology of the century as an item on a CEO’s wish list.
6. The Constitutional Question Nobody Wants to Ask
Inside the Capitol, a different conversation has been gathering force for months — one that the AI order intensifies rather than answers. It is the conversation about whether the President of the United States is, in the constitutional sense, capable of doing the job.
“Unable to discharge the powers and duties of his office.” The framers left the word unable undefined — on purpose.
he 25th Amendment, ratified in 1967, contains in its Section 4 the only constitutional mechanism for involuntarily removing a sitting president who is still alive and in office. The text is famously, deliberately spare: it provides for transfer of power when the Vice President and a majority of the Cabinet (or “such other body as Congress may by law provide”) declare in writing that the President is “unable to discharge the powers and duties of his office.”
The word the entire mechanism turns on — unable — is not defined anywhere in the Constitution. The Bayh–Celler drafters of the amendment chose that ambiguity. They knew that “inability” could not be reduced to a list. A president can be incapacitated by surgery, by stroke, by dementia, by paranoia, by drug interaction, by ideological capture, by sheer chaos of judgment. The framers declined to pre-write the medical chart. They left the question of fitness to the political branches and the moment.
The Argument Being Made
That ambiguity is precisely what Rep. Jamie Raskin (D-MD), ranking member of the House Judiciary Committee and a constitutional law professor, invoked in April when he introduced legislation to create a 17-member Commission on Presidential Capacity to Discharge the Powers and Duties of the Office. The commission — a body Congress is explicitly authorized to establish by Section 4’s “such other body” clause — would be empowered to medically evaluate the president and report its findings. More than fifty Democrats co-sponsored. More than seventy publicly called for Trump’s removal that month, after the president posted on Truth Social that “a whole civilization will die tonight” if Iran did not meet his demands.
Rep. Raja Krishnamoorthi (D-IL) put it as plainly as it can be put, in a statement on April 7, 2026: the Iran threat was “not an isolated incident, but part of a dangerous pattern of reckless escalation, erratic decision-making, and general conduct that raises grave questions about his fitness to discharge the duties of the presidency.”
Where the AI Order Fits
The AI episode is not a single dramatic incident. It is a slow-motion documentation of the pattern Raskin and Krishnamoorthi described. A president who repeals critical national-security guardrails because his predecessor signed them. Who is then persuaded eighteen months later, by his own intelligence officials, that he was wrong. Who schedules a signing ceremony, invites executives, then cancels at the last hour because he “didn’t like certain aspects” of an order he himself was about to sign. Who allows three billionaires to rewrite the document in twelve days. None of this, in isolation, is incapacity. All of it, together, is what “unable to discharge” looks like when the inability is one of judgment, focus, and continuity rather than consciousness.
The Practical Barriers — Honestly Stated
The path to invocation is, in 2026, effectively closed. Section 4 requires Vice President JD Vance and a majority of Cabinet secretaries — every one of them a Trump loyalty appointee — to initiate the process. They will not. Even if they did, a two-thirds supermajority of both chambers of Congress would be required to sustain removal over Trump’s objection, and the current Congress is Republican-controlled. Raskin’s commission bill will not pass this Congress and Trump would veto it if it did. As CNN reported last week, Republicans have dismissed 25th Amendment talk outright.
But political impossibility is not the same thing as constitutional irrelevance. The framers of the 25th Amendment did not build the mechanism for easy political weather. They built it for moments when the country needs a way to say, in the constitutional record, that something is wrong — even if the wrong cannot, yet, be corrected. The moral and constitutional case for naming Trump’s pattern of conduct as a question of capacity is not weakened by the absence of votes in the Senate. It is strengthened by the fact that the absence of those votes is itself part of the institutional failure the framers feared.
7. The Bottom Line
The Tuesday AI executive order will be remembered, in the histories that get written, as a small thing. A short document, voluntarily complied with by a handful of companies, mostly forgotten by the next news cycle. That is the wrong way to read it. The order is small precisely because the presidency that produced it is too erratic to produce anything large. Frontier AI is the most consequential technological transformation since electricity. The American response to it has been written by a man who cannot keep his own signing ceremonies on the calendar.
The country deserves, at minimum, a national AI policy that is the product of deliberation rather than late-night phone calls from venture capitalists. It deserves a Congress that does its constitutional job and writes durable law. It deserves a press corps that names the pattern rather than chronicling each fresh installment of it. And it deserves, above all else, a president who is present — in every sense of that word — to the moment he was elected to govern.
Editorial Conclusion
The President of the United States has just spent eighteen months proving he could not be trusted with the most consequential technology of our lifetimes — and then, in a private ceremony with the ink barely dry on a draft his own donors rewrote, asked the country to call it leadership.
It is not leadership. It is the wreckage of leadership being labeled as the thing itself. The 25th Amendment exists because the framers understood that a presidency can be incapacitated by chaos as surely as by coma. The path to invoking it is closed today; the case for the case grows every week it remains closed. What is required of citizens is not despair. What is required is to keep the constitutional record — to name what is happening, with sources and on the page, so that when the political weather turns, the country knows exactly what it lived through.
America has had enough. The Constitution has waited long enough. The work is to make sure the record is clear when the rest of the country is ready to read it.
Sources & References
- CNBCTrump signs AI executive order asking companies to give government early access to models
- NBC NewsTrump signs order seeking early access to powerful AI models before release
- CNN BusinessTrump signs executive order asking for access to new AI models before they launch
- The Washington PostTrump signs order designed to give government early look at powerful AI models
- CBS NewsTrump signs AI executive order to give government early look at new models
- NPRTrump’s new AI safety order seeks voluntary review of new models
- PBS NewsHourTrump signs executive order that allows voluntary federal vetting of top AI models
- CyberScoopTrump administration releases scaled-back AI executive order
- Cybersecurity DiveTrump signs EO seeking early government access to powerful AI models
- Council on Foreign RelationsAssessing Trump’s Executive Order on AI Oversight
- Sen. Mark Warner (D-VA)Warner statement on Trump’s AI Executive Order
- Sen. Edward Markey (D-MA)Statement on Trump AI Regulation Moratorium Executive Order
- The HillRep. Raskin introduces bill to assess president’s fitness under 25th Amendment
- Rep. Raja Krishnamoorthi (D-IL)Krishnamoorthi Calls for President Trump’s Removal Under 25th Amendment
- CNN PoliticsWhite House breaks from precedent by not releasing Trump’s medical report
- TechCrunchPresident Trump repeals Biden’s AI executive order
- The RegisterFrontier AI safety tests may be creating the very risks they’re meant to stop
- AI FrontiersThe Hidden AI Frontier — on model-weight theft and leak risk
Related News
