How Elon Musk, Mark Zuckerberg, and a Single Night of Lobbying Killed Trump’s AI Executive Order
In the high-stakes world of artificial intelligence regulation, the line between industry influence and public policy is often blurred. But rarely has it been erased so completely—and so quickly—as it was on a single night in late March. What was supposed to be a ceremonial signing of a new AI executive order by President Donald Trump instead became a last-minute cancellation, driven by direct phone calls from two of the most powerful figures in tech: Elon Musk and Mark Zuckerberg.
The backstory, first reported by Semafor, reveals a startlingly candid account of how the Trump administration’s tentative steps toward AI oversight were halted not by internal debate or legislative gridlock, but by a coordinated push from the very CEOs whose companies would have been most affected. This is not a story about a sweeping regulatory crackdown that never happened. It is a story about a modest, voluntary framework that still proved too much for an industry that has grown accustomed to writing its own rules.
The Executive Order That Almost Was: What Trump Scrapped
Before we dissect the politics, it is essential to understand precisely what was on the table—and what was not. The executive order that Trump had planned to sign was not the kind of heavy-handed regulation that tech companies typically warn against. There was no licensing regime. There were no mandatory hold periods. There were no criminal penalties for non-compliance.
Instead, the proposed order would have established a voluntary mechanism for AI developers to engage with federal agencies. Under this framework, companies like OpenAI, Google DeepMind, Meta, and xAI would have been encouraged—not required—to submit advanced AI models for security review up to 90 days before their public release. Think of it as a pre-flight check: the government would get an early look at potentially powerful systems, offer feedback, and flag any obvious safety concerns. The company would then decide whether to incorporate that feedback or proceed without it.
That was it. Voluntary. Non-binding. A handshake agreement with a federal seal of approval. And it was still deemed unacceptable.
Why the White House Walked Away
Trump’s public explanation was characteristically blunt. “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,” he told reporters in the Oval Office on Thursday. He added that he had postponed the order “because I didn’t like certain aspects of it,” declining to specify which ones. His main concern, he said, was that the order “could have been a blocker.”
That language is telling. For a president who has consistently framed AI as both a jobs engine and a national security imperative, even the perception of a “blocker” was politically toxic. The phrase suggests that the administration’s internal calculus had shifted from cautious governance to all-out acceleration—a worldview that sees any regulatory friction, no matter how light, as a competitive disadvantage against China.
But Trump’s public framing left out a crucial detail: the order had been effectively killed by the very industry it was meant to oversee.
The Night the CEOs Called: Musk, Zuckerberg, and Sacks in the Room
According to Semafor and corroborating US media reports, the critical intervention happened between Wednesday night and Thursday morning. Over the course of that single evening, three individuals spoke directly with President Trump:
- Elon Musk, CEO of xAI and Tesla, and a self-described “free speech absolutist” who has repeatedly warned against AI regulation.
- Mark Zuckerberg, CEO of Meta, whose company has invested billions in open-source AI models like Llama.
- David Sacks, the venture capitalist who, until recently, served as Trump’s AI and cryptocurrency tsar, and who retains significant influence within the administration.
The argument these three made was not about the technical details of the order. Instead, they appealed to what insiders call the “accelerationist” faction within the administration—a group that includes officials at the National Economic Council and staffers in the Vice President’s office. The core message was simple: any review mechanism, even a voluntary one, creates precedent. Precedent leads to expectations. Expectations lead to demands. And demands, over time, become mandates.
In other words, the CEOs argued that the executive order was not a one-off gesture but a camel’s nose under the tent. If the government can ask to see your models today, they suggested, it can demand to approve them tomorrow.
Why This Argument Landed
The accelerationist worldview is not without its merits. The United States is in a high-stakes race with China for AI supremacy. Beijing has poured state resources into AI development, with companies like Baidu, Alibaba, and ByteDance receiving direct government support. Any regulatory friction in the US could, in theory, slow domestic innovation and hand the advantage to Chinese firms that operate under far looser constraints.
But the flaw in this reasoning is that it treats all regulation as equally harmful. A voluntary, non-binding review mechanism is not the same as a licensing board with the power to shut down projects. Yet the CEOs’ framing conflated the two, and the accelerationist faction was receptive. The result: a policy that had been in development for months, delayed multiple times, was scrapped in a matter of hours.
The Broader Regulatory Vacuum: No Law, No Framework, No Accountability
This episode did not occur in a vacuum—pun intended. The United States currently has no comprehensive federal AI legislation. What governance architecture exists has been assembled piecemeal, through executive orders, agency guidance, and voluntary agreements. The result is a patchwork of inconsistent rules that vary by sector, by state, and by administration.
Consider the timeline:
- 2020: The Trump administration issued its first executive order on AI, emphasizing trust and innovation.
- 2022: The Biden administration released a Blueprint for an AI Bill of Rights, a non-binding set of principles.
- 2023: The White House secured voluntary commitments from major AI developers to prioritize safety testing—a pledge with no enforcement mechanism.
- 2025 (March): The federal Centre for AI Standards and Innovation announced evaluation agreements with Google DeepMind, Microsoft, and xAI, allowing the government to assess models before public availability.
That last initiative is notable because it continues regardless of Thursday’s non-signing. The Centre for AI Standards and Innovation still has agreements in place with the three companies. But those agreements are bilateral and limited in scope. They do not cover the broader industry, and they do not create any systematic process for transparency.
The States Step In, The Feds Step Back
As the federal government stalls, individual states are moving forward. States like California, Colorado, and Connecticut have introduced or passed their own AI legislation, covering everything from algorithmic bias to deepfake disclosure. This creates a compliance nightmare for companies that operate nationwide. A model that is legal in Texas might require additional disclosures in Illinois. The patchwork is inefficient, but it is better than nothing.
In early March, the Trump administration released a National AI Legislative Framework that urged Congress to preempt state-level AI laws that “impose undue” burdens. The irony is thick: the same administration that killed a voluntary federal framework is now asking Congress to block state-level initiatives that are more binding. The message is clear: the federal government wants AI governance, but only on its own terms—and only if industry agrees.
What This Means for the AI Industry and Its Users
For non-engineers trying to make sense of this, the practical implications are significant. Let’s break them down:
1. Unchecked Model Release
Without even a voluntary pre-release review, companies can launch AI models with no external oversight. This is not necessarily catastrophic—most companies do internal testing—but it means safety concerns are addressed behind closed doors, not in public view. If a model has a dangerous vulnerability or a tendency to generate harmful content, the public may not find out until after the damage is done.
2. The Accelerationist Bet
The Trump administration is betting that removing all regulatory friction will accelerate US innovation and outpace China. This is a high-risk strategy. Regulation can be heavy-handed, but it can also serve as a forcing function for safety research, disclosure, and public trust. In its absence, trust erodes, and the backlash—when it comes—may be harsher than any voluntary framework could have been.
3. Industry Capture
The lobbying victory achieved by Musk, Zuckerberg, and Sacks is a textbook example of industry influence over policy. When three CEOs can directly reverse a presidential executive order through overnight phone calls, the line between public interest and private interest becomes dangerously thin. This is not to say that the CEOs are acting in bad faith—they genuinely believe that regulation slows innovation. But their incentives are not aligned with broader societal safety.
4. A Missed Opportunity
The executive order that was scrapped was modest by any standard. It did not require anything. It did not threaten anyone. It simply created a channel for communication and a timeline for review. By killing it, the administration sent a signal that even minimal government engagement is unwelcome. That signal will be heard not just in Silicon Valley, but in Brussels, Beijing, and every other capital where AI governance is being shaped.
The Path Forward: What Comes Next?
With the executive order dead, the question is what—if anything—will replace it. The administration’s National AI Legislative Framework suggests a preference for federal preemption of state laws, but Congress shows no signs of passing comprehensive legislation. The 118th Congress has held hearings on AI but has not advanced any major bills.
Meanwhile, the Centre for AI Standards and Innovation will continue its voluntary evaluation agreements with a handful of companies. But these agreements are narrow, and they do not cover the full ecosystem of open-source models, startups, and foreign developers that now populate the AI landscape.
The irony of the accelerationist approach is that it may ultimately produce the opposite of its intended effect. By refusing to engage in even minimal governance, the US creates regulatory uncertainty. Investors hate uncertainty. If companies do not know what rules they will face in two years, they may slow their investments or shift development to jurisdictions with clearer frameworks, such as the European Union, which is implementing the AI Act.
In the end, the CEOs got what they wanted: no executive order. But they may have also gotten what they did not want: a fragmented, unpredictable regulatory environment that makes long-term planning impossible. The race with China continues, but without a governance framework, it is a race run in the dark.