President Donald Trump signed an executive order Tuesday establishing a voluntary framework for the federal government to review the most advanced artificial intelligence systems for national security risks before they reach the public, a move that came less than two weeks after he shelved an earlier version over concerns it could slow American innovation.
The order gives the government up to 30 days to vet an AI system flagged for review, with the process led by the director of the National Security Agency. Participation by AI developers is voluntary, a distinction the White House was quick to emphasize.
The signing carried no ceremony. Trump had canceled an Oval Office event with tech industry executives last month after objecting to language in a previous draft. He declined to sign an earlier version on May 21, worried that heavy-handed oversight could dull America's technological edge against China and other competitors.
The executive order itself frames the challenge plainly. "Advanced AI capabilities make our Nation stronger, but also introduce new national security considerations that require coordinated action across executive departments and agencies," it states.
Trump put the competitive stakes in blunter terms:
"We're leading China, we're leading everybody, and I don't want to do anything that's going to get in the way of that lead."
That tension, between maintaining America's AI dominance and ensuring new systems do not introduce catastrophic vulnerabilities, is what delayed the order in the first place. The fact that Trump pulled the plug on the May 21 signing and then reworked the text before issuing it without fanfare suggests the White House took the calibration seriously.
The White House posted on social media Tuesday that the order "creates a process for frontier labs to voluntarily share cutting-edge cyber models in order to secure critical infrastructure and strengthen the government's own cyber defenses." The post added, in emphatic capitals, "We are NOT conducting oversight of all new models, as that level of government overreach would have chilling effects on free speech and innovation."
The backdrop to the order is a specific AI system that rattled senior officials. Anthropic announced its most advanced model, Claude Mythos, in April. The system's apparent ability to identify cybersecurity vulnerabilities in software prompted an urgent response from the highest levels of the financial establishment.
Treasury Secretary Scott Bessent and outgoing Federal Reserve Chair Jerome Powell convened an emergency meeting with Wall Street CEOs soon after the Mythos announcement. The two officials warned bank executives about the risks posed by Mythos' capabilities.
Anthropic has limited access to Mythos to a small group of trusted partners, including major tech companies and banks. On Tuesday, the company said it expanded that group by another 150 organizations.
That a single AI model could trigger a joint Treasury-Fed warning to the nation's largest financial institutions tells you something about the speed at which this technology is outrunning the government's ability to keep up. The executive order is, in part, an attempt to close that gap, without creating a regulatory chokepoint that drives AI development offshore or underground.
The major AI companies lined up behind the order, though their statements carried the careful phrasing of firms that want Washington's blessing without Washington's hand on the steering wheel.
Anthropic called the order "an important step in strengthening America's leadership in AI" and said it looks forward to collaborating with the White House on implementation. That endorsement is notable given that Anthropic is simultaneously locked in a legal fight with the Trump administration over a contract dispute with the Pentagon.
OpenAI's chief global affairs officer, Chris Lehane, offered a statement that read more like a position paper than a celebration:
"As AI capabilities continue to advance, we believe 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."
Google also described the policy as an important step.
The industry's willingness to participate in a voluntary review process, rather than fight it, suggests the leading AI firms recognize that some form of pre-release national security screening was coming whether they liked it or not. Better to shape it now than have Congress impose something heavier later.
Juan Londoño, a policy analyst at the libertarian-leaning Cato Institute, called the order "a step in the right direction to prepare the nation for the release of advanced AI systems." He applauded the White House for making the process voluntary.
But Londoño flagged real concerns about vagueness in the order's implementation. He questioned how the government, specifically the NSA director, will decide which AI models qualify for scrutiny and which "trusted partners" get early access to reviewed systems. Without clear criteria, Londoño warned, the framework could set a "dangerous precedent" that a future administration might "weaponize" against disfavored companies or technologies.
That concern deserves attention. A voluntary process under a president who understands the value of American tech dominance is one thing. The same architecture in the hands of an administration hostile to the industry, or eager to pick winners and losers, is something else entirely.
The order also represents a course correction of sorts. Trump repealed many of former President Joe Biden's AI guardrails just hours after returning to the White House last year, viewing them as bureaucratic drag on innovation. Now, roughly a year and a half later, the administration has crafted its own framework, one that prioritizes national security while keeping participation voluntary and capping the review window at 30 days.
Democratic Sen. Mark Warner, vice chairman of the Senate Intelligence Committee, welcomed the policy but could not resist a jab. He said the administration had "belatedly discovered the need to redo something it hastily dismantled in its first year."
Warner's criticism misses the mark. Repealing Biden's approach and then building a different one is not the same as rebuilding what was torn down. Biden's framework leaned toward broad regulatory control. Trump's order is narrower, voluntary, and time-limited, a fundamentally different philosophy. The question was never whether AI needed national security review. The question was whether that review should function as a gate or a guardrail.
For all the order accomplishes, significant gaps persist. The criteria for determining which AI models trigger federal scrutiny remain undefined in public reporting. The process for selecting "trusted partners" who receive early access to vetted systems is unclear. And the extent to which the final order differs from the version Trump rejected on May 21 has not been disclosed.
These are not minor details. They are the hinges on which the entire framework swings. A 30-day voluntary review sounds reasonable in principle. Whether it stays reasonable depends entirely on how the NSA and participating agencies define the boundaries.
Congress will likely have its own ideas. Warner's committee has been circling AI oversight for months. The executive order may buy the administration time, but it will not settle the legislative debate over how much authority the federal government should have over frontier AI systems.
Trump's instinct to protect American AI leadership from bureaucratic suffocation was correct when he repealed Biden's rules. His instinct to ensure the most powerful AI systems get a national security look before deployment is also correct. The challenge is holding both truths at once, and this order attempts to do exactly that.
The voluntary structure is the key. It keeps the government in a consultative role rather than a gatekeeping one. It preserves the competitive incentive for American companies to build fast and build first. And it signals to allies and adversaries alike that the United States takes AI security seriously without outsourcing innovation decisions to a federal review board.
When a single AI model can spook the Treasury Secretary and the Fed Chair badly enough to summon Wall Street's top executives for an emergency briefing, the technology has outgrown the "move fast and break things" era. The question is whether Washington can keep pace without becoming the thing that breaks.