Three Phone Calls Killed the AI Order

Introduction

On May 20, Anthropic confirmed that its newest model could find zero-day software vulnerabilities in the systems banks and hospitals run on. By May 21, the White House had a 7-page draft executive order asking AI labs to share frontier models with the federal government for a voluntary 90-day review before public release. By May 22, three phone calls from Elon Musk, Mark Zuckerberg, and David Sacks had ended it. A Future of Life Institute poll released the same week showed 79% of Republican voters supported exactly the kind of pre-release review those calls killed.

What the Order Actually Said

The draft executive order Politico obtained was the lightest possible AI safety framework. Across 7 pages, it asked labs to volunteer their frontier models to the Treasury Department, the NSA, and the Office of the National Cyber Director for review at least 90 days before public release, and it created a repository for tracking security flaws.

The order also contained a sentence written specifically to prevent the thing Sacks would later publicly claim it would become: "Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models." The drafters wrote the slippery-slope argument out of the document before anyone could make it. Industry still pushed to shrink the review window from 90 days to 14, and some unnamed firms wanted it cut entirely. By the planned signing ceremony, the White House had landed on the most voluntary framework it could legally write, and three people with direct financial exposure picked up the phone.

What the Three Callers Stand to Lose

Start with xAI's government book. In July 2025, the Pentagon's Chief Digital and Artificial Intelligence Office awarded Musk's company a contract with a floor of $2 million and a ceiling of $200 million for frontier AI work across national security applications, alongside Anthropic, Google, and OpenAI. NBC News reported that xAI was a "late addition" to the program, slotted in after March 2025 under the Trump administration, despite a former Pentagon AI lead saying there hadn't been "a single discussion with anyone from X or xAI" during the original planning. Senate Minority Leader Chuck Schumer called the contract "wrong" and "dangerous" on the Senate floor, citing Grok's documented antisemitism incident.

Two months later, in September 2025, GSA and xAI announced a OneGov agreement making Grok 4 and Grok 4 Fast available to every federal agency at $0.42 per organization for 18 months, valid through approximately March 2027. GSA itself called it "the longest term for a OneGov AI agreement to date." In December, the War Department agreed to embed Grok models in GenAI.mil, the bespoke government AI platform with planned access for 3 million military and civilian personnel at the IL5 security classification.

Sacks's exposure is wider. The NYT's November 2025 investigation of his financial disclosures found he holds 708 total tech investments, of which at least 449 are in companies with ties to AI. NPR confirmed that even after divesting some positions (Amazon, Meta, Musk's xAI), Craft Ventures still maintains "more than 400 investments in tech firms with ties to AI." He took the White House AI/crypto czar role as a "special government employee," a designation that exempted him from confirmation hearings and full public disclosure. He left that role in late March 2026 but kept attending administration briefings. Earlier the week of the cancellation, he sat in on briefings about this specific executive order and reportedly indicated he wouldn't oppose the voluntary review. Then he picked up the phone.

Meta is the odd one out on direct contracts. USAspending shows Meta's federal exposure at $7,000, a DEA purchase order from June 2025. Zuckerberg's interest is competitive: a 90-day window in front of Llama or any next-generation closed model is a 90-day product delay against OpenAI, Anthropic, and xAI.

The Calls

The reporting is consistent across three independent newsrooms. Semafor broke the story on May 21 with this lead: "The Trump administration's plans for an executive order regulating artificial intelligence were put on hold this week after some of the tech industry's biggest players, including Elon Musk, Mark Zuckerberg, and venture capitalist David Sacks, persuaded the White House to call it off, according to people familiar with the matter." The same piece reports the lobbying succeeded "because Musk and others were able to appeal to the accelerationist crowd, including officials at the National Economic Council."

The Washington Post followed with four bylines and three independent sources at 1:41 AM on May 22, describing the specific sequence: Trump made his decision following conversations with Sacks first, then spoke to Musk and Zuckerberg by phone Thursday afternoon, then told staff the order could not proceed. Politico corroborated and obtained the draft text. The Guardian, Fortune, Axios, Reuters, and AFP all confirmed the same outline.

The argument Sacks brought to Trump, per Politico's senior White House official: the voluntary reviews "may one day become mandatory" — the specific interpretation the EO's statutory language had been written to foreclose. Sacks built the public version on his All-In podcast in the days before the call, framing pre-release review as part of a "doomer industrial complex" of Effective Altruists, former Biden staffers, and Anthropic trying to "create a permanent new infrastructure in Washington" — the infrastructure the order he killed had already foreclosed.

The Denial

After the Washington Post story published, Musk posted on X: "This is false. I still don't know what was in that executive order and the president only spoke to me after declining to sign." Meta issued a parallel statement saying Zuckerberg had spoken to Trump only after the order was rescinded.

Semafor and Politico directly attribute the cancellation to Musk's lobbying. Semafor names him as part of the group that "persuaded the White House to call it off." Politico's reporting describes calls to Trump from Musk that derailed the signing. The Washington Post placed those calls before, not after, Trump told staff the order could not proceed.

The sequence dispute is one thing. The other half of Musk's post is the part I keep rereading: "I still don't know what was in that executive order." Politico obtained the draft and published it. Axios published its contents two days before the scheduled signing. Tech executives at other major labs were being briefed on the terms, which is how the reporting on industry's 14-day counter-proposal exists in the first place. And Musk runs a company sitting on a $200 million DoD ceiling and an 18-month all-agencies GSA agreement, both of which the review framework would have touched directly. The denial went out after the WaPo story was already public, which means he was calling a piece of reporting "false" that he had already read.

Who Benefits

xAI keeps shipping frontier models into the Pentagon's CDAO program, the GSA all-agencies pipeline, and the War Department's GenAI.mil platform with no 90-day buffer in front of any of them. Meta keeps its product launch calendar clean against OpenAI and Anthropic. Sacks's 400-plus AI-tied portfolio at Craft Ventures avoids a federal framework that, even with its explicit ban on mandatory licensing, would have set a precedent of pre-release government review across the whole frontier.

None of these three faced a regulatory comment period, a Congressional vote, or a rulemaking docket where their objections would land in a public record. The process for killing this one was three phone calls with no transcript and no required disclosure. Sacks was a private citizen by May 21 (he'd left his White House role two months earlier), and he still attended the official briefings, and he still called the president. The accelerationist faction inside the National Economic Council, per Semafor's sourcing, provided the internal conduit. No transcript exists, no list of who was on which call, just the outcome.

The Coalition Gap

The Future of Life Institute poll released May 22 found 79% of Republican voters support government testing of AI before release. 87% support government power to block models posing national security threats. A separate ARI poll found 71% of Republicans prefer mandatory testing versus 17% who favor voluntary-only. Economist/YouGov polling shows 68% of Republicans say AI is advancing too fast.

The political tell landed three days before the cancellation. On May 18, more than 60 "America First" leaders (including Steve Bannon, Amy Kremer, and Brendan Steinhauser) signed an open letter organized by Humans First urging Trump to require mandatory testing and government approval of frontier AI models. Bannon, the loudest voice in America First, asked for mandatory testing — more than the draft EO's voluntary review even contemplated. Three days later, the call sheet had Musk, Zuckerberg, and Sacks on it — none of whom had signed the letter.

Three Calls, One Precedent

The most voluntary version of pre-release review that could legally be written, with statutory text foreclosing any reading as mandatory licensing, was killed by three phone calls from three people with documented financial interests in killing it.

On May 21, the Bannon letter and the FLI poll both ran into the same wall: a private citizen with 449 AI investments calling from outside the building, and an internal White House faction with no public record of what got said. If a voluntary 90-day review with explicit anti-licensing language can't survive a single afternoon of billionaire phone calls, the next AI safety framework won't be judged on what it says. It'll be judged on which three people get the call sheet and which government conduit takes the call.