Trump signs narrower AI executive order with voluntary 30-day pre-release reviews

President Donald Trump signed an executive order on June 2, 2026 that asks advanced AI developers to voluntarily submit new models to the federal government for testing 30 days before public release. The final order is significantly narrower than earlier drafts and explicitly forbids any mandatory licensing of frontier AI, settling weeks of negotiations between the White House and an industry that warned a heavier-handed regime would stall U.S. competitiveness.

What's new

• 30-day voluntary pre-release window. Covered AI companies are asked to share new models with government evaluators 30 days before deploying them to the public. A previous draft had set the window at 90 days; industry insiders had pushed for closer to two weeks.
• No mandatory licensing. The order states that nothing in it authorizes "the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models."
• AI-assisted crime as a DOJ priority. The directive instructs the Department of Justice to treat AI-assisted offenses such as hacking as high-priority enforcement areas.
• Cybersecurity tooling access. It also calls for making it easier for federal, state, and local agencies and operators of critical infrastructure to access cybersecurity tools embedded in frontier AI models.

Context

Trump was originally slated to sign a more demanding version of the order in late May but pulled back after pushback from senior figures inside the administration's AI orbit, including venture capitalist and former White House AI czar David Sacks. The Bloomberg reporting that anticipated the final shape of the order in early May framed it as a U.S. AI-security directive that would stop short of mandatory model tests — a posture the signed text now confirms.

The move arrives in the same week the White House published a separate consolidated AI policy agenda and as OpenAI released its own "blueprint for democratic governance of frontier AI," both of which lean on voluntary commitments rather than statutory regulation. The administration's preference for voluntary pre-release review echoes the existing U.S. AI Safety Institute model evaluation arrangements first established under the prior administration, rather than the licensing regime laid out by the EU AI Act's Code of Practice for General Purpose AI.

Why it matters

The order sets the operating baseline for how Washington plans to relate to frontier model developers for at least the next budget cycle: closer cooperation, no permitting gate, and a relatively short review window that is unlikely to delay shipping schedules. For developers, the practical consequence is a new 30-day handoff to add to existing safety testing workflows; for the policy community, it confirms that the appetite in the U.S. executive branch for binding model-level rules has, for now, collapsed.

The DOJ enforcement language is the more durable hook. By naming AI-assisted hacking specifically as high-priority, the order signals that the near-term federal response to frontier-model risk will be reactive — prosecuting misuse — rather than preventative through permitting. That posture aligns Washington's stance more closely with the cybersecurity-first framing that Anthropic, OpenAI, and Microsoft have all pushed in recent public-policy positions, and further from the European model of pre-deployment conformity assessment.