Musk and Zuckerberg Called Trump. His AI Safety Order Was Dead in Minutes.

📖 5 min read

On Thursday, May 21, 2026, the White House had a signing ceremony scheduled. Invitations had gone out. AI company executives had been briefed. Then Elon Musk, Mark Zuckerberg, and David Sacks picked up their phones – and within minutes, the whole thing was cancelled.

President Trump abruptly postponed signing a landmark executive order on AI oversight, just hours before the planned signing. The order would have required AI companies to share their most advanced models with the federal government up to 90 days before public release. Tech’s most powerful figures made a few calls, and the policy was shelved.

What the Order Would Have Done

The executive order, months in the making, had three core requirements for AI companies building “frontier models” – the most powerful AI systems at the cutting edge:

• Engage with the government before releasing covered models
• Provide government access to those models at least 90 days before public launch
• Share access to select critical infrastructure providers ahead of release

The framework was technically “voluntary,” but the 90-day pre-launch review would have given agencies including the NSA the ability to test models for dangerous capabilities and identify vulnerabilities. For fast-moving AI labs, a mandatory 3-month government review window before every major model launch is anything but voluntary in practice – it fundamentally changes the release timeline.

The order had bipartisan support among AI safety advocates, and even some AI companies had been cooperating with the planning process. A photo-op signing ceremony had been scheduled at the White House for Thursday afternoon.

The Phone Calls That Killed It

According to reporting from the Washington Post, Axios, and Politico, Elon Musk, Mark Zuckerberg, and former White House AI and crypto czar David Sacks called Trump in the hours before the scheduled signing. The argument they made: a pre-launch review process would slow US AI innovation and hand China a strategic advantage in the AI race.

Sacks, who previously served as Trump’s AI policy point man before leaving the role earlier in 2026, reportedly “hated” the order. One source familiar with the situation told Axios that the main reason for the delay was simple: Trump “just hates regulation.”

Trump himself, speaking to reporters, said he postponed the order because he “didn’t like certain aspects of it” – without specifying which aspects. The White House has not announced a revised timeline for when or whether the order will be signed in any form.

Why This Matters Beyond Washington

This isn’t just a political story. The decision has direct consequences for how quickly AI develops – and how much anyone outside the companies building it knows about what they’re releasing.

What the EO Would Have Meant	What Happens Without It
90-day government review before launch	No mandatory pre-launch review
NSA access to test frontier models	No federal security testing required
Slower release timelines for top labs	Labs set their own timelines
Government visibility into AI capabilities	Public learns about capabilities at launch
Potential safety catch before public exposure	Market and public become the test

The US currently has no binding requirements for AI safety testing before release. The EU’s AI Act – which does have mandatory requirements for high-risk AI systems – is being phased in through 2027. The gap between the world’s two largest AI markets on regulation is now wider than it has been since AI became a strategic priority.

The Argument Against the Order

The tech industry’s concerns weren’t entirely manufactured. A 90-day mandatory review window would have meaningfully affected competitive dynamics. OpenAI, Anthropic, Google DeepMind, and Meta all ship major model updates on cadences measured in weeks to months. Adding a 3-month government gate to every frontier release would compress available market windows and create bureaucratic bottlenecks that smaller labs – with less capacity to navigate government review – would feel disproportionately.

The China argument is also not trivial: Chinese AI labs including DeepSeek, Baidu, and Alibaba operate under no equivalent constraint, and the gap in release velocity could have real competitive consequences if US labs were forced into extended review cycles.

That said, “voluntary” frameworks have a poor track record of actually constraining behavior when commercial incentives point the other way. The same companies arguing against government review have signed various voluntary AI safety commitments over the past two years with limited measurable impact on their development timelines.

The Optics Problem

Regardless of the merits, the optics of this episode are striking. The three men who most directly benefit from AI companies facing zero pre-launch regulatory friction called the President, and within minutes a government safety policy was killed. Musk owns xAI. Zuckerberg runs Meta AI. Sacks is a major AI investor and former government AI official.

That’s not a conspiracy – that’s how lobbying works. But it does illustrate the structural dynamic now governing AI policy in Washington: the people best positioned to influence AI regulation are the people with the most to lose from it. No equivalent phone bank exists for the interests of people who might be harmed by powerful AI systems released without review.

What Happens Next

The White House has not announced a revised order or a new timeline. The executive order that was killed was itself a weaker version of earlier proposals – the “voluntary” framing had already been added in response to industry pushback before the signing ceremony was ever scheduled. It is unclear whether a further-revised version will emerge, or whether the US government will attempt any pre-launch AI oversight mechanism in the near term.

Congress has several AI-related bills in various stages of progress, but none are close to a floor vote in either chamber. The EU AI Act’s mandatory provisions remain the only binding framework for any major AI lab with European market access.

BetOnAI Verdict

The story here isn’t that tech executives lobbied against a regulation. That’s normal. The story is how fast it worked and how little resistance it met.

The executive order had legitimate flaws – the 90-day review window was blunt, the “voluntary” framing was already a compromise that may not have achieved much, and the China competitiveness concern is real. A better-designed framework would be worth arguing for rather than mourning this one.

But the net result is the same: the US moves toward more powerful AI with less mandatory transparency than the alternative. If you believe frontier AI systems carry real risks that companies won’t catch on their own, Thursday was a bad day. If you believe speed matters most and the market will punish dangerous AI, it was a good day.

What’s not in dispute: the rules of AI development are being written right now, and the people writing them are not waiting for government to catch up.

---

Sources:

• Washington Post – Last-minute lobbying by tech industry officials led Trump to cancel AI order
• Axios – Why Trump’s AI executive order was pulled
• Politico – Trump postpones AI executive order
• Axios – Scoop: Trump AI executive order seeks early government access to frontier models
• New York Times – Trump Cancels Signing of Executive Order Granting Oversight of A.I. Models