Three Phone Calls Killed Trump's AI Safety Order: Musk, Zuckerberg, and Sacks Convince President to Scrap Voluntary 90-Day Review
Overnight lobbying by Elon Musk, Mark Zuckerberg, and David Sacks persuaded President Trump to cancel a planned executive order that would have let federal agencies preview frontier AI models up to 90 days before release.
Overview
President Trump called off the signing of a draft executive order on artificial intelligence on May 22, 2026, hours before the ceremony was scheduled to take place. The decision came after Elon Musk, Mark Zuckerberg, and former AI and cryptocurrency adviser David Sacks spoke directly with Trump between Wednesday night and Thursday morning, according to Axios. The cancellation leaves the United States without a formal pre-deployment review process for frontier AI models, even a voluntary one, and has set off a scramble inside the administration over what, if anything, comes next, as Axios reported.
What the Order Would Have Done
The draft executive order would have established a voluntary system allowing AI companies to submit advanced models to national security agencies for testing and vetting up to 90 days before public release, according to Fortune. It required no licensing regime and set no mandatory hold periods.
Under the order, Treasury, the NSA, and CISA would have jointly assessed AI models’ advanced cyber capabilities and defined thresholds for which systems would qualify as covered frontier models, the Washington Times reported. CISA would have received authority to issue binding directives for civilian government cyber defense. The Defense Secretary would have been required to prioritize military cyber defense, and the Treasury Secretary would have led a new voluntary body coordinating software vulnerability scanning with industry. The order also targeted federal systems, critical infrastructure operators, state and local governments, and partnerships with AI developers.
The Office of the National Cyber Director was granted two months under the draft text to develop the evaluation processes, and the order included the creation of a reporting vault for AI-related security vulnerabilities, NewsOne noted.
Despite its scope on paper, AI News described the order as “lightweight” and “non-binding” in regulatory terms — a characterization that made its cancellation all the more striking to observers of AI governance.
The Lobbying Campaign
Sacks, Musk, and Zuckerberg communicated with Trump between the night of May 20 and the morning of May 21, before the planned signing, Axios reported. Sacks, who until recently served as Trump’s AI and cryptocurrency adviser, argued that the voluntary framework would function as de facto licensing, slowing AI releases, according to Fortune. All three framed the proposal as regulation incompatible with US competitiveness, AI News reported.
The lobbying operated through allies at the National Economic Council and the vice president’s office, bypassing standard regulatory channels, according to AI News. The order never reached public comment or congressional review before being scrapped.
An anonymous source close to the discussions told Axios that the order was “just something doomers wanted,” using language that signals how the accelerationist faction within the administration characterizes AI safety advocates.
Trump’s Reasoning
Trump confirmed the cancellation to reporters, saying “I didn’t like certain aspects of it,” without elaborating on the specific provisions he objected to. He framed the decision in competitive terms: “I think it gets in the way of — you know, we’re leading China, we’re leading everybody, and I didn’t want to do anything to get in the way of that lead,” Axios reported. He added that he believed the order “could have been a blocker” and said “I really thought that could have been a blocker” and “I want to make sure that it’s not,” according to NewsOne.
A source familiar with the discussions told Axios that Trump “just hates regulation.”
A Split With the MAGA Base
The cancellation creates an unusual gap between the administration’s AI policy and the preferences of its core voters. According to Future of Life Institute polling cited by Fortune, 79 percent of Republicans favored government testing of AI models for safety, and 87 percent supported the government having the power to block unsafe releases.
More than 60 MAGA loyalists, including Steve Bannon, signed an open letter urging model testing before release, and conservative group Humans First also backed oversight, Fortune reported. Anthropic and OpenAI had also indicated support for a voluntary vetting framework before the signing was cancelled.
What Comes Next
The White House’s Office of the National Cyber Director indicated it is working on additional AI security initiatives, though no timeline has been given, Axios reported. Federal officials told NewsOne that the order would be revised rather than abandoned entirely.
In the absence of a formal framework, the administration is applying selective scrutiny to individual models through ad hoc arrangements, including a program called Project Glasswing focused on Anthropic’s Mythos model, while applying less rigorous monitoring to the distribution of OpenAI’s GPT-5.5, Fortune reported.
The episode arrives as the federal government’s Centre for AI Standards and Innovation — CAISI — continues its pre-deployment evaluation agreements with Google DeepMind, Microsoft, and xAI, AI News noted. Those voluntary agreements represent the closest thing the US now has to a structured pre-release review process, operating outside the executive order framework that was cancelled.
The Machine Herald previously reported on the infighting inside the administration that had already slowed the order’s development before the final lobbying effort killed it entirely. With the order off the table, key questions remain unanswered on the future of government access to top AI models and general AI safety, Axios noted.