President Trump postponed signing an executive order on May 21 that would have required AI companies to submit new models for government review before public release. The order, drafted in response to escalating concerns about frontier model capabilities, would have given national security and civilian agencies up to 90 days to evaluate models before they reached users.
“I didn’t like certain aspects of it, I postponed it,” Trump told reporters in the Oval Office, according to UPI. “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 the Order Would Have Done
The executive order represented a reversal of the administration’s previous hands-off approach to AI regulation. According to PYMNTS, citing The Wall Street Journal, the order would have established two mechanisms:
First, national security and cyber officials would have worked with agency leaders and AI companies to address software vulnerabilities identified by increasingly powerful models. Second, the Treasury Department would have led a voluntary clearinghouse to find and fix security flaws in unreleased models.
The order’s stated goal was protecting smaller institutions, including community hospitals and banks, that lack the resources for sophisticated cyber defenses, PYMNTS reported.
The Mythos Catalyst
The push for oversight accelerated after Anthropic released its Mythos model, which can identify software vulnerabilities at an advanced level. Administration officials and business leaders flagged the model because its capability to discover zero-day exploits at scale could also enable cyberattacks, according to PYMNTS. OpenAI is previewing a similar model to customers and, like Anthropic, has limited access with government input.
Anthropic’s own Project Glasswing initiative, launched after Mythos autonomously discovered thousands of high-severity zero-day vulnerabilities, underscores why regulators felt urgency. The tension between that urgency and the competitiveness argument is what ultimately stalled the order.
Last-Minute Collapse
The signing ceremony fell apart partly due to logistics. Leaders of OpenAI, Google, Anthropic, Meta, and Microsoft were invited to the White House just 24 hours before the event, according to UPI, citing the New York Times. The rushed invitations caused cancellations that annoyed the president.
Vice President JD Vance had briefed stakeholders on May 19 about balancing safety and innovation, according to Politico. But Trump’s public comments suggest the balance tipped toward industry freedom. He described government model review as something that “could have been a blocker.”
The Competitiveness Calculus
The delay leaves the U.S. without a federal framework for pre-release AI model review at a moment when frontier models are demonstrating capabilities that regulators in 2024 would have classified as hypothetical. Anthropic’s Mythos finds real vulnerabilities. OpenAI’s reasoning models solve previously intractable mathematical problems. The models arriving next quarter will be more capable still.
For agent builders and platform operators, the postponement preserves the current status quo: no mandatory government review before deploying new foundation models. That means faster access to new capabilities, but also no federal safety net when those capabilities surface in autonomous agent systems operating at scale.
The White House said the signing would be rescheduled. No new date has been announced.