President Donald Trump signed an executive order Tuesday inviting developers of advanced artificial-intelligence systems to submit their models to the federal government for as much as 30 days of cybersecurity vetting before public release, formalizing a voluntary review regime that stops short of any licensing requirement.

The order, titled “Promoting Advanced Artificial Intelligence Innovation and Security,” defines categories such as “covered frontier model” and “advanced cyber capabilities,” but goes out of its way to disclaim coercive authority. “Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models,” the text reads.

Trump signed in private, according to CNBC. The setting was conspicuous given what preceded it: a planned White House ceremony on May 21 that the president scrapped at the last moment, telling reporters he “didn’t like certain aspects of it.” The Washington Post noted it wasn’t immediately clear how Tuesday’s version differs from the one he declined to sign less than two weeks earlier.

One difference is documented. An earlier draft contemplated a 90-day review window that the president pulled over concerns it would blunt U.S. competitiveness, according to a Council on Foreign Relations analysis published the same day. The final order cuts that to 30. The arc, from 90 days to 30 to optional, traces the gravitational pull of the competitiveness frame inside this administration’s tech policy.

The substantive case for federal vetting isn’t hypothetical. CFR’s analysis cites Anthropic’s Claude Mythos among models that have shown the ability to “autonomously identify and exploit hidden vulnerabilities in real-world software,” and points to documentation from Google’s threat intelligence team of state-aligned actors already using frontier models to automate cyberattacks. The capabilities the order gestures at are already deployed in the wild.

The Anthropic backdrop complicates the optics. The Department of Defense recently designated the company a supply chain risk, barring defense contractors from using its technology; Anthropic has sued to reverse the designation, and that litigation is ongoing. A voluntary review program arrives in the middle of an active federal-versus-lab dispute over the most capable systems on the market.

Whether labs will actually participate is the open question. CFR’s read is that they probably will, “if only to forestall more invasive regulation later.” That’s not safety culture. It’s the same calculation the financial sector made around stress tests after 2008: accept the legible federal touch now to avoid the illegible one later.

Sources