On the morning of May 21, 2026, Donald Trump was scheduled to sign an executive order that would have given the US government up to 90 days of access to frontier AI models before public release.

By afternoon, the event was cancelled.

What Happened

The order—drafted after Anthropic flagged cybersecurity risks in its Mythos model—would have expanded voluntary government testing of the most capable AI systems. The goal was straightforward: identify vulnerabilities before release, patch critical infrastructure, and create a coordinated response to AI-enabled threats against banks, hospitals, and utilities.

Trump pulled the plug after learning that some CEOs couldn't attend the signing ceremony. According to the New York Times, he'd given them 24 hours' notice. Others who rearranged their schedules were "midair on their way to the Oval Office" when they found out the trip was for nothing.

But the scheduling snub wasn't the whole story.

Who Killed It

Multiple reports point to a coordinated push from Trump's own allies:

  • David Sacks, Trump's former AI czar, reportedly called the president directly on the morning of May 21 to warn the order would slow innovation and hurt the US in its AI race with China. Sacks had previously warned that "AI ideologues or doomers" were trying to use Mythos to "create a permanent new infrastructure in Washington."

  • Elon Musk denied involvement, posting on X: "This is false. I still don't know what was in that executive order." Meta said Mark Zuckerberg spoke to Trump only after the order was already cancelled.

  • OpenAI, notably, supported the order according to Semafor—perhaps sensing an opportunity to lock in a regulatory framework that favors incumbents who can afford compliance.

The Stakes

The collapse leaves the US with no formal plan for managing risks from the most powerful AI systems. Trump scrapped Biden's 2023 AI oversight order on his first day back in office. This was the replacement—and it lasted less than a day.

The gap is stark: the EU's AI Act entered force in 2024 with binding rules for high-risk systems. China has its own framework. The US has nothing.

Industry pushed for a 14-day pre-release testing window instead of 90. Even that compromise couldn't survive a phone call from Silicon Valley.

What This Means

For AI labs, the message is clear: in this administration, lobbying beats regulation. The companies best at playing politics—not necessarily the ones building the safest systems—will set the rules.

For everyone else: the guardrails aren't coming. At least not from Washington.

Sources: Ars Technica, The New York Times, Semafor, Reuters, The Straits Times