Summary
On June 2, 2026, President Trump signed an executive order titled "Promoting Advanced Artificial Intelligence Innovation and Security," a modified version of a draft order he cancelled on May 21. The order establishes a voluntary framework under which AI companies may submit frontier models for government security testing up to 30 days before public release, and directs the Treasury Secretary to form an AI cybersecurity clearinghouse within 30 days in consultation with the NSA Director, the CISA Director, and the National Cyber Director. No pre-release review is mandatory. The signed order differs substantially from the cancelled draft, most notably in reducing the maximum voluntary review window from 90 to 30 days and shifting clearinghouse authority from NIST to the Treasury Department.
What Happened
On May 21, 2026, President Trump cancelled a planned signing ceremony for an AI cybersecurity executive order hours before it was to take effect, citing concerns that the order "could have been a blocker" to U.S. AI competitiveness. The draft had included a 90-day voluntary pre-release review window and designated CISA and NIST as the agencies responsible for model evaluations. White House AI adviser David Sacks, along with Elon Musk and Mark Zuckerberg — executives with direct financial stakes in the outcome — had spoken with Trump the night before the cancellation. Axios published the full text of the cancelled draft on May 22.
The signed order, released on June 2 under the title "Promoting Advanced Artificial Intelligence Innovation and Security," preserves the core structure of the cancelled draft but with significant modifications. The review window was reduced from 90 days to 30 days. The designated evaluating agencies shifted from NIST to a Treasury-led body: the order directs the Secretary of the Treasury, acting in consultation with the National Cyber Director, the NSA Director, and the CISA Director, to form an AI cybersecurity clearinghouse within 30 days of the order's signing.
The clearinghouse is tasked with coordinating and deconflicting scanning for software vulnerabilities, discovering and validating such vulnerabilities, and coordinating remediation and distribution of patches across government and industry participants on a voluntary basis.
The order directs federal agencies to develop benchmarks assessing the cybersecurity capabilities of AI models — both offensive and defensive — and instructs agencies to use AI tools to strengthen the government's own security posture. The order's preamble explicitly frames its cybersecurity provisions as a response to strategic competition with China. The entire pre-release review mechanism is voluntary: AI companies that choose not to participate face no legal penalty, and the order creates no mandatory disclosure obligations.
Why It Matters
The June 2 signing closes a 12-day gap opened by the May 21 cancellation and establishes the first formal mechanism under the Trump administration for the U.S. government to access frontier AI models before they reach the public — however optionally. The 30-day voluntary window represents the administration's negotiated position between the industry-opposed 90-day draft and no structured access at all. Its practical effect will depend entirely on which labs participate, on what terms, and on what the government does with the access it receives.
The clearinghouse architecture — Treasury-led, voluntary, focused on vulnerability coordination rather than model capabilities evaluation — reflects the administration's stated preference for public-private partnership over regulatory mandates. The shift from NIST to Treasury as the lead agency is not incidental: NIST has an established AI safety framework (the AI RMF, updated in 2024) and is institutionally associated with measurement and standards; Treasury is primarily a financial regulator with no prior AI technical evaluation role. The choice signals that the clearinghouse is conceived as an industry coordination body rather than a technical evaluation function.
The order does not establish mandatory pre-release review of any kind. The March 2026 National AI Legislative Framework called for federal preemption of state AI laws but created no federal review requirements; this order also does not. The gap between the U.S. approach — voluntary, framed around innovation and competition with China — and the EU's parallel regulatory trajectory continues to widen. In the same week the Trump order was signed, the EU was finalizing the AI Act Omnibus amendments, which extended compliance deadlines for high-risk AI systems while preserving the mandatory review architecture established in 2024.
§ How to read the metadata
- Landmark
- Fundamentally alters the trajectory; 2–5 per year.
- Major
- Meaningfully shifts the landscape; 2–4 per month.
- Notable
- Worth documenting; significance can be upgraded later.
- Confidence
- High = primary sources corroborate. Medium = credible secondary only. Low = provisional. Disputed = credible sources disagree.
- Contestation
- Uncontested = no formal challenge. Contested = at least one challenge open. Superseded = replaced by a later entry. Unresolved = dispute still open.