Over the past two weeks, apparent leaks have given us a peek inside the drafting of a Trump Administration executive order aimed at creating safeguards around frontier AI models. The latest report, which is from Axios, says the most significant of those safeguards may now be voluntary.
The order initially sounded like it was going to be a fairly sweeping change that would have put a government agency, possibly the federal government’s Center for AI Standards and Innovation (CAISI), in charge of vetting all new models.
It was a significant change from an earlier Trump Administration policy document released about two months ago, which had given the impression that the Trump Administration was determined to be lax in the extreme on AI regulation—essentially calling for nothing more onerous than age restrictions for users. The Administration’s overall perspective on AI had seemingly been laid out in a 2025 speech at the AI Action Summit in France by Vice President J.D. Vance, which more or less renounced regulation completely.
But further apparent leaks, and other strange, subtle moves around the Trump Administration and AI have suggested that the drafting of this order may not be going smoothly. Now, with Axios’ latest release of anonymously sourced information, it sounds like there’s been a bit of a reversal.
The latest version of the order is apparently divided into two sections: “cybersecurity” and “covered frontier models.” The cybersecurity section reportedly bolsters federal cybersecurity measures—not so much regulating AI as hardening infrastructure against the perceived AI threat.
The frontier models section, however, would establish a “voluntary framework.”
Makers of frontier models would apparently have a 90 day window to check in with the government and have their model vetted. But if it’s voluntary, that doesn’t make much business sense.
Read the full article here
