← Back to feed

The White House had the tech executives lined up for a signing ceremony Thursday afternoon. The order never got signed.

President Trump pulled back an executive order establishing a voluntary pre-release oversight framework for frontier AI on May 21, hours before a scheduled ceremony at the Oval Office. He told reporters he had not liked what he had seen in the order’s text. “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.”

What he saw was this: a seven-page draft, obtained in full by Politico, that would have created a 90-day voluntary review window before AI companies release their most powerful models. Participating labs — Anthropic, OpenAI, Google, and others — would submit systems to federal evaluation in advance of public launch. The draft repeated three times that reviews would remain voluntary. “Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement.”

The assurances were not enough.

How It Was Killed

David Sacks placed a call to Trump on the morning of May 21. Sacks, who had served as Trump’s AI and crypto czar before stepping down, argued the draft’s voluntary review framework would create the architecture for an FDA-style approval process for AI. His argument: voluntary frameworks have a history of becoming mandatory once the regulatory infrastructure is in place, and any review process that touches frontier models gives the government a lever it will eventually use.

Elon Musk, whose xAI is one of the primary targets of any such review, also contacted the White House before the ceremony. Mark Zuckerberg — whose Meta is the largest open-weight model provider and therefore most exposed to review obligations — reached out separately. The three interventions, according to reporting by Cryptobriefing and the AP, persuaded Trump to halt the signing.

The administration has not said when or whether the order will be revised and rescheduled. JD Vance told reporters Tuesday that the administration wants to promote innovation while addressing cybersecurity threats and data privacy. “The president wants us to be pro-innovation. He wants us to win the AI race against all other countries in the world.” He did not address the postponement directly.

What the Order Would Have Done

The draft gave federal agencies — the Commerce Department, DHS, and intelligence bodies — access to frontier models 90 days before public release for evaluation. Participating companies would receive threat intelligence in exchange. The attorney general would enforce the Computer Fraud and Abuse Act against anyone using AI for illegal access.

Nothing in the draft created binding obligations on any company. The framework was designed to look like an extension of existing voluntary information-sharing programs at the FBI and CISA. The Commerce Department had already signed voluntary evaluation agreements with Google, Microsoft, and xAI — agreements that later disappeared from the department’s website, without explanation.

The Internal Division

The postponement exposed a structural fault line inside Trump’s coalition. National security advisers and some Republican legislators have pushed for tighter oversight of frontier AI systems, particularly after Anthropic’s Claude Mythos Preview demonstrated it could autonomously chain zero-day exploits across major operating systems. The Project Glasswing revelations — 10,000-plus vulnerabilities identified in 47 days — put the AI security question directly in front of policymakers.

Silicon Valley aligned advisers, led by Sacks and backed by Musk and Zuckerberg, hold the opposite view: that any pre-release process creates a compliance burden that Chinese labs operating without equivalent constraints will exploit. In their framing, the risk of competitive disadvantage outweighs the risk of misuse.

Trump aligned himself with the deregulation camp. Former White House AI policy lead Dean Ball described the divisions as “healthy tension” and said he would welcome an order focused on government AI cooperation — “but ultimately, I’m fine with them taking time to get this right.”

What Happens Next

Without the order, the current posture holds. A handful of voluntary evaluation agreements exist between NIST and frontier labs. The Commerce agreements with Google, Microsoft, and xAI are in force but not publicly visible. Anthropic and OpenAI have their own existing agreements with government bodies.

The AI National Legislative Framework, which the White House released earlier this month, pushes for federal preemption of state AI laws. That document does not include pre-release oversight requirements. If the EO is revised, its voluntary review architecture may survive; what Sacks argued against was any structure that could later justify mandatory controls. Whether the next draft addresses that distinction remains to be seen.