Trump Signs Narrowed AI Executive Order Establishing Voluntary Pre-Release Review Framework

President Donald Trump signed an executive order June 3 directing federal agencies to establish a voluntary framework for government review of advanced AI models before public release, ending weeks of internal White House conflict over the scope of federal oversight of frontier artificial intelligence systems. The order asks AI developers to voluntarily submit their most powerful "frontier models" to the federal government for security review up to 30 days before release. The order directs federal agencies, including the Defense Department, the Treasury, and the Cybersecurity and Infrastructure Security Agency, to shore up the country's cybersecurity defenses for critical infrastructure. Trump signed the order without a public ceremony.

The order is a materially narrower instrument than the version the administration drafted in May. The administration originally scheduled a signing ceremony for May 21, invited tech CEOs to attend, and held a senior-official briefing on the morning of the planned announcement, but the signing was canceled at the last minute. Trump told reporters after scrapping the earlier directive that it "could have been a blocker" to AI innovation, adding, "We're leading China, we're leading everybody, and I don't want to do anything that's going to get in the way of that lead." The signed order stipulates that the voluntary testing mechanism would give the federal government access to cutting-edge AI models "up to 30 days before they plan to release such models to other trusted partners," compared with a 90-day window in the original draft. Industry executives had lobbied for a review window closer to two weeks; the 30-day compromise represents a landing point between that preference and the 90-day timeline national security officials had sought.

The order contains an explicit constraint on regulatory creep. The text states that "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." The NSA director, however, retains broad discretion to determine which AI models qualify for scrutiny and which "trusted partners" receive early access to them, a provision that has drawn scrutiny from civil liberties advocates. Separately, the order directs federal agencies to develop benchmarks to assess AI models' cyber capabilities and to create an "AI cybersecurity clearinghouse" to review and share vulnerability information. That clearinghouse, to be stood up within 30 days, will be coordinated by the Treasury Secretary, the National Cyber Director, the NSA, and CISA. The order does not invoke a specific statutory authority for the review framework itself; it operates under the president's general Article II authority over executive branch agencies and national security policy.

The executive action arrives in direct response to Anthropic's release of its Claude Mythos Preview model in April. The model sent policy ripples through Washington with its ability to find critical vulnerabilities in widely used operating systems; experts characterized it as a significant leap in cybersecurity capabilities. Mythos Preview is a general-purpose frontier model that, according to Anthropic, can surpass all but the most skilled humans at finding and exploiting software vulnerabilities, and has already identified thousands of high-severity flaws, including in every major operating system and web browser. The order represents a reversal of the administration's prior posture: on his first day in office in 2025, Trump rescinded a Biden administration requirement for AI companies to submit safety tests to the government, characterizing that mandate as overly burdensome. Mythos's debut upended that calculus, forcing the administration to reconsider the government's role in evaluating rapidly evolving AI risks.

On the same day the order was signed, Anthropic separately announced a significant expansion of Project Glasswing, the controlled-access coalition it formed to deploy Mythos Preview for defensive cybersecurity purposes. Following several weeks of collaboration with Project Glasswing partners, the security industry, open-source software maintainers, and the U.S. government, Anthropic is extending the partnership to approximately 150 new organizations. The new organizations, based in more than 15 countries, include infrastructure operators in sectors not previously represented in the coalition, such as power, water, healthcare, and telecommunications, as well as hardware vendors, critical software maintainers, and nonprofit groups. New partners must meet Anthropic's security requirements before gaining access to the model. Project Glasswing's existing cohort has already surfaced more than 10,000 high- or critical-severity software vulnerabilities since the program launched in early April. Anthropic has stated it plans to expand the program even further, prioritizing additional essential infrastructure providers, maintainers of critical open-source software, and safety testers.

The two developments together mark a structural inflection point in how the executive branch is positioning itself relative to frontier AI. The voluntary framework the president signed stops short of the regulatory posture of analogues abroad. The European Union's AI Act, entering full enforcement in August, provides a direct contrast: it establishes mandatory requirements, statutory authority, and penalties for non-compliance. Domestically, the order creates institutional architecture, including the clearinghouse and benchmark-development process, that a future administration could convert into binding authority through subsequent rulemaking or legislation. The order was signed while Anthropic remains in the middle of a legal fight with the Trump administration over a contract dispute with the Pentagon, a posture that underscores the extent to which commercial litigation, national security policy, and AI governance have become intertwined in this regulatory cycle.