Trump’s Light AI Touch Keeps Getting Heavier

The Donald Trump Experience reopened in January 2025, with the new administration vowing, contra the Biden White House, to govern artificial intelligence with the lightest of touches. It was a great start. The president’s executive order on Day One: “ Removing Barriers to American Leadership in Artificial Intelligence .”  That was then, this is now. The administration’s decision last Friday to slap export controls on frontier Anthropic’s Fable and Mythos models—which the company met by pulling both offline within the hour—is the latest action suggesting the previous “do no harm” impulse is no longer operational. Whether this change of direction should comfort those who favor a firmer hand is another question.  Populist governance has a signature : an impatience with rules, commitments, and institutions that constrain what populist leaders want to do today. That, especially given their role as embodying the will of the good and wise People.  L’affaire Anthropic  bears every mark of this tendency. The new order barred foreign nationals from the two models and offered no public explanation of the specific threat. The company said it had been given 90 minutes to pull the models, with no details on the actual threat—a deadline, not a negotiation. Officials disputed that characterization, saying they had spent hours urging Anthropic to act voluntarily before smashing the export control button.  Worse, the move capped a run of interventions, including a designation of Anthropic as a supply-chain risk to national security. Then there are the president’s repeated musings about the government taking equity stakes in leading AI companies. The light touch gets heavier with each news cycle.  Trump’s June 2 executive order had promised a voluntary 30-day pre-release notification process—thus explicitly rejecting mandatory licensing. Friday showed how fast that promise buckles once national security enters the conversation. The real-world execution here amounts to licensing by other means, and the administration’s fight with Anthropic could reasonably be interpreted as implying that every major model requires government permission to ship.  Tech policy analysts Adam Thierer and Neil Chilson, no friends of heavy regulation, argue  that durable and predictable rules are needed, ASAP. They want Congress to pass a statutory framework for frontier-model safety, with clear standards and civilian oversight. America’s AI lead, they warn, will erode under the sort of ad hoc governance now on display.  As I concluded  in my newsletter essay yesterday:  To me, it’s an inescapable conclusion that the restraint light-touch advocates have counted on is expiring. The forces of tighter-grip regulation look to be holding the upper hand, and I fear the only open question now is just how heavily Washington decides to lean on the sector. A formal “FDA for AI” prerelease licensing regime, given the trajectory, is no longer unthinkable.  A contrary case worth noting comes from  Tom Davidson of the Forethought think tank, who reads the same chaos as a blow to the case for intervention itself. Suppose Washington decided it wanted to slow frontier AI—an actual pause of some kind coordinated with China, the world’s other leading AI power. Good luck with that, as Davidson argues in this X post:  Perhaps the resulting new consensus across Washington will be that trying to micro-regulate or pause a fast-evolving technology is a fool’s errand. There are certainly worthier ideas for consideration . My guess is that those who want such interventionist policies will merely conclude that they—not the former advocates of AI laissez-faire—are the ones to implement them. The post Trump’s Light AI Touch Keeps Getting Heavier appeared first on American Enterprise Institute - AEI .