Commentary

Trump signs AI executive order creating voluntary 30-day model review process — critics call it a licensing regime in disguise

Jun 2, 2026

Key Points

  • Trump's AI executive order establishes a voluntary 30-day model review process, down from an original 90-day proposal abandoned after regulatory pushback.
  • The evaluation criteria remain classified, creating a de facto licensing regime where companies cannot know what standards they must meet before deployment.
  • Dean Ball argues the structure trades transparency for appearance of due diligence, giving government deniability while researchers lack clarity on regulatory thresholds.

Summary

Trump's AI Executive Order: 30-Day Review Masquerades as Voluntary

President Trump signed an AI executive order today creating a voluntary 30-day pre-deployment model review process with the U.S. government. The move represents a significant retreat from an earlier draft that proposed a 90-day review window — a draft the administration abandoned after pushback over regulatory overreach.

The order allows AI companies to submit new models for government evaluation before deployment. The government will respond within 30 days with commentary or advice. The evaluation criteria remain classified.

The licensing regime subtext

Dean Ball, founder of Hyperdimensional, reads the order as laying groundwork for a model licensing regime disguised as a voluntary process. The classification of evaluation rubrics is the tell. No company knows what it's being graded on.

This mirrors how the FDA operates with low-risk products: silence functions as tacit approval. A company submits a model, receives no objection within 30 days, and treats that silence as a clear signal to proceed. The structure gives the government deniability while creating de facto licensing infrastructure.

Ball calls it a "fairly major win for the safety contingent within the admin and a significant loss for the SACS accelerationist wing." He is skeptical nonetheless.

The classification problem

The details of how the voluntary system will work — what tests run, what benchmarks matter, what triggers escalation — are classified. This creates a mismatch between transparency and enforcement. Most lab staff lack security clearances. If the regulatory thresholds themselves are classified, researchers training models won't know whether they're triggering the review process until they've already crossed the line.

Ball frames the frustration in concrete terms: it's like taking a driver's license test on a mystery rubric. You show up, they test you on criteria you can't see, and you don't know if you passed or failed until weeks later.

What the intelligence community plans to accomplish in 30 days is unclear. Penetration testing and adversarial red-teaming are possible. So is designing attack vectors around bioweapons capability. All are defensible uses of time. None require public explanation.

Ball's final assessment: the order is a mistake — not catastrophic, but a durable one that trades clarity for the appearance of due diligence.

Every deal, every interview. 5 minutes.

TBPN Digest delivers summaries of the latest fundraises, interviews and tech news from TBPN, every weekday.