Answer-First Lead
President Trump cancelled a planned executive order on AI safety testing on Thursday, May 22, after last-minute phone calls from Elon Musk, Mark Zuckerberg, and former AI advisor David Sacks. Executives from major tech companies were already en route to the White House for the signing ceremony when it was abruptly called off. The US now has no formal system for reviewing frontier AI models before release.
🔍 THE BOTTOM LINE
The most powerful tech executives in America successfully lobbied to kill a voluntary AI safety review system — one that wouldn’t have been mandatory — just as models like Anthropic’s Mythos demonstrate they can find 10,000+ security vulnerabilities faster than humans can patch them.
What happened
The White House had drafted an executive order creating a voluntary vetting system for frontier AI models. Under the proposed order, AI companies would submit their models to federal agencies up to 90 days before release so the government could test for dangerous capabilities and find weaknesses before hackers or foreign actors could exploit them, according to the Washington Post.
The draft explicitly ruled out mandatory government licensing or pre-approval. It was, by all accounts, a modest proposal — a voluntary system that companies could opt into. The push came in response to new models like Anthropic’s Mythos, which can independently find and exploit security flaws in code, the Washington Post reported.
The last-minute lobbying
On Wednesday evening and Thursday morning, three phone calls changed everything:
- Elon Musk (SpaceX CEO) called Trump to warn the review system could slow AI development
- Mark Zuckerberg (Meta founder) raised similar concerns about competitive disadvantage
- David Sacks, Trump’s former AI and crypto advisor, called Trump directly on Thursday morning — without telling his own staff — and derailed the order, Politico reported, citing a senior White House official
According to The Decoder, Sacks had been briefed on the draft by science advisor Michael Kratsios, staff secretary Will Scharf, and National Cyber Director Sean Cairncross. He initially indicated he could live with it. Then, late Wednesday night, he started raising concerns that the voluntary system could become mandatory in practice and be abused by future administrations.
By Thursday morning, Sacks called Trump directly and killed it.
The internal rift
The cancellation exposed deep divisions within the administration:
| Faction | Position |
|---|---|
| Commerce Department + OSTP | Light-touch, pro-industry approach |
| Office of National Cyber Director | Security-focused, wants governance now |
| David Sacks | Voluntary systems risk becoming mandatory; anti-regulation |
| JD Vance | Now publicly prioritizing data/privacy protection after Mythos concerns |
A government official told Axios the effort was “something the so-called doomers wanted” — a term used dismissively in tech circles for voices warning about existential AI risks.
The order was also divisive within the industry itself. OpenAI lobbyist Chris Lehane broadly supported it, according to Politico. Other companies pushed to cut the 90-day review window to just 14 days. The draft’s plan to give the Treasury Department a lead role raised eyebrows, since safety reviews are typically handled by CISA and NIST.
The power vacuum
Sacks’ tenure as AI advisor ended recently, creating what The Information described as a “power vacuum” within the White House’s AI leadership structure. Sacks continues to visit the White House weekly but no longer holds an official position.
Meanwhile, Trump agreed at the recent summit with Chinese President Xi Jinping to “launch an intergovernmental dialogue on AI” to mutually navigate emerging national security risks, the Chinese Foreign Ministry confirmed, according to Ars Technica. China’s State Council has already outlined a legislative work plan for 2026 to “improve AI governance and accelerate comprehensive legislation for the sound development of AI.”
What is the US-China AI governance gap? While the US has no formal frontier model review system after this cancellation, China is accelerating regulatory processes. Beijing issued regulations in April requiring domestic AI firms to establish internal “artificial intelligence ethics review committees,” and the State Council has outlined a 2026 legislative work plan for AI governance.
Why it matters
This isn’t just process politics. The timing matters:
- Anthropic’s Mythos found 10,000+ high/critical vulnerabilities in its first month under Project Glasswing, with only 97 patched
- Cloudflare flagged 2,000 bugs (400 high/critical); Mozilla found 271 Firefox vulnerabilities (10x previous model); Palo Alto shipped 5x the normal patch volume
- The order was drafted in direct response to these capabilities — and was killed the same week they were demonstrated at scale
- JD Vance, who publicly prioritized “protecting people’s data” and “people’s privacy” after Mythos concerns, now sits in an administration that just cancelled the only mechanism for reviewing these models
What is the voluntary vetting system? It was a proposed executive order framework where AI companies would voluntarily submit frontier models to federal agencies up to 90 days before release for safety testing. The system was explicitly not mandatory and ruled out government licensing or pre-approval — but critics argued voluntary systems could become de facto mandatory in practice.
What happens next
The White House says it will revisit the order, but no timeline has been set. In the meantime:
- The US has no formal process for reviewing frontier AI models before deployment
- China is moving faster on AI governance legislation
- The intelligence community is simultaneously pushing to expand access to frontier models like Mythos for classified work
- The Pentagon continues to blacklist Anthropic as a supply-chain risk while other agencies scramble to use Claude
The collision between “move fast” tech billionaires and “slow down” safety advocates has been playing out for years. This week, the fast side won — and they didn’t even need to show up to the signing ceremony.
❓ Frequently Asked Questions
Q: Was the cancelled order mandatory? No. The proposed executive order created a voluntary system where companies could submit frontier models for government review up to 90 days before release. It explicitly ruled out mandatory licensing or pre-approval.
Q: Who lobbied against it? Elon Musk (SpaceX), Mark Zuckerberg (Meta), and David Sacks (former White House AI advisor) made direct calls to Trump. Sacks, who had initially indicated support, reversed course Wednesday night and called Trump Thursday morning without informing his own staff, according to Politico.
Q: What does this mean for AI safety? The US currently has no formal mechanism for reviewing frontier AI models before deployment. Models like Anthropic’s Mythos can find 10,000+ security vulnerabilities in a month, but there’s no government process for evaluating those capabilities before release.
Q: What about NZ? New Zealand has no domestic frontier model regulation either. The AI Forum’s refreshed “Blueprint for Aotearoa” through 2030 emphasises voluntary adoption frameworks rather than mandatory review, making the US decision a relevant signal for NZ’s own governance trajectory.
Q: Is China regulating AI faster than the US? By some measures, yes. Beijing now requires domestic AI firms to maintain ethics review committees, and the State Council has outlined 2026 legislation for AI governance. The US cancelled its voluntary review system this week.
