Musk and Zuckerberg phone calls cut Trump AI review from 90 to 30 days

President Trump signed an on June 2, 2026 titled "Promoting Advanced Artificial Intelligence Innovation and Security." The order asks AI companies to voluntarily submit frontier models to the federal government for security review up to 30 days before public release. Participating companies include Anthropic, OpenAI, and Google.

The signed order is a scaled-back version of an earlier draft. That draft called for a 90-day review window and was planned for signing on May 21, 2026. Trump cancelled the May 21 ceremony after three phone calls from Elon Musk, Mark Zuckerberg, and David Sacks the night before.

Musk, Zuckerberg, and Sacks each called Trump directly, bypassing , congressional input, and any public deliberative process. They argued the original draft's 90-day window would slow American AI development relative to China and could establish a precedent for de facto government gatekeeping of new models.

The final signed order represents a 67% reduction in the review window, from 90 days to 30 days, and converts any review obligation from mandatory to voluntary. Industry executives had actually pushed for a two-week window; the 30-day compromise landed between that preference and the national security officials' 90-day demand.

All three callers carry direct financial interests in minimizing AI oversight. Elon Musk owns xAI, the company behind the Grok chatbot, which was expected to go public at approximately $1.75 trillion through SpaceX. Zuckerberg's Meta develops Llama, its open-source AI model series, which competes directly with proprietary models subject to safety review requirements. Sacks co-founded Craft Ventures and holds broad AI startup investments across the sector.

David Sacks served as White House AI and Crypto Czar from January 2025 until March 2026, when he stepped down after hitting the 130-day service limit for special government employees. Trump then appointed him co-chair of the President's Council of Advisors on Science and Technology (PCAST). Sacks specifically told Trump that the voluntary review provision would become a de facto licensing regime, and that a future administration could easily convert it to a mandatory requirement.

The final order explicitly states that nothing in the executive order "shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement" for new AI models. Legal analysts at WilmerHale noted the anti-mandatory-licensing language is a direct response to industry and marks a notable shift from the administration's prior deregulatory stance, while still falling far short of binding government authority.

OpenAI publicly supported the original 90-day review framework, with Chief Global Affairs Officer Chris Lehane stating the company believes "effective safety frameworks should continue to be developed through democratic institutions." That position put OpenAI in direct opposition to Musk's xAI and Zuckerberg's Meta on the oversight question. The split is notable: Musk is currently in federal litigation against OpenAI over allegations that it abandoned its nonprofit mission.

Anthropic withheld its Mythos Preview model in April 2026 after finding the model could autonomously identify and exploit software vulnerabilities across major operating systems. That announcement triggered alarms inside the administration and provided the national security rationale for pursuing any pre-release review mechanism.

The order directs federal agencies to develop benchmarks assessing AI models' cyber capabilities and create an "AI cybersecurity clearinghouse" to review and share information on vulnerabilities. Within 60 days of signing, the NSA and CISA must develop a classified benchmarking process to determine what models qualify as "covered frontier models" subject to voluntary review.

The Council on Foreign Relations assessed the order as an attempt to engineer a cybersecurity window: giving government defenders preferential access to frontier AI capabilities while attempting to delay adversary access through a prerelease evaluation period. Critics note that without mandatory participation, the window functions only if companies choose to cooperate.

is the mechanism on display. A 2025 academic analysis found AI companies deploy tactics including private meetings with regulators, claims that regulation stifles innovation, and lobbying pressure to ensure oversight remains favorable. What happened on the night of May 20, 2026 skipped even those institutional channels: three executives with direct financial stakes in the outcome called the president directly, and got a 67% reduction in the review window plus full voluntariness enshrined in law.

Congress has never passed a comprehensive AI governance law. The United States regulates AI entirely through executive orders, agency guidance, and state laws — all of which companies can challenge in court or outlast through a change in administration. Biden's EO 14110 lasted 15 months before Trump rescinded it on January 20, 2025.

The AI industry registered 3,570 lobbyists in 2025 — more than one lobbyist for every member of Congress. Thirty-seven states had AI-related legislation pending as of early 2026. The night of May 20, 2026, three executives bypassed all of it with direct presidential phone calls.

The corporate-capture pattern in AI mirrors a documented playbook. Reagan's 1981 executive order requiring cost-benefit analysis for all major regulations gave industry a durable lever to delay agency action. The 1996 Telecommunications Act, written largely by telecom lobbyists, deregulated broadband and created the monopoly conditions that persist today. Citizens United v. FEC (2010) removed limits on corporate political spending, accelerating the consolidation of corporate influence over regulatory policy. The AI industry has compressed this cycle: from sector emergence to direct presidential access in under a decade.