President Donald Trump’s last‑minute decision to halt an artificial intelligence safety order has exposed a sharp split between those pressing for tougher oversight of frontier AI models and those warning against slowing U.S. innovation.

Early plans for an AI safety framework

In mid‑May, the White House prepared a long‑anticipated executive order aimed at cybersecurity and “covered frontier models,” including cyber‑capable systems like Anthropic’s Mythos.¹ The draft, later obtained by reporters, outlined a voluntary framework in which AI labs would give the government early access to advanced models—up to 90 days before public release—for security review.² It stressed that “nothing” in the text would create mandatory licensing or pre‑clearance for new AI models.²

Proponents saw this as a modest first step to “preventing a ‘Chernobyl moment’ in AI,” arguing that structured testing of frontier systems was urgently needed to head off catastrophic misuse.³

Rising concerns and industry pushback

As details emerged, Axios reported that the order sought early government access to frontier models while bolstering cybersecurity for national security agencies and critical infrastructure.¹ But industry briefings revealed growing unease: companies worried that 90‑day review windows could delay launches or force design changes, undermining their competitive edge.⁴

Behind the scenes, Trump adviser David Sacks and several tech executives argued the plan was “unnecessary” and driven by AI “doomers,” reflecting the president’s instinctive dislike of regulation.⁵

A photo‑op collapses and the order stalls

By May 21, the White House had scheduled a signing ceremony featuring top AI and tech CEOs, but the event unraveled just hours before it was to begin.⁵ Some leading CEOs, including from major AI firms, declined to attend on short notice, prompting frustration in the West Wing and contributing to Trump’s decision to pull the plug.⁴⁶

Trump publicly said he “didn’t like certain aspects” of the order and worried it “could have been a blocker” to jobs and the “tremendous good” AI is creating, emphasizing that the U.S. is “leading China” and he did not want to jeopardize that lead.⁶⁷

Competing visions for AI governance

Safety advocates, pointing to models capable of quickly finding and exploiting security flaws, argue that structured testing and clearer rules are essential despite the order’s voluntary design.³⁴ Accelerationist voices in Trump’s orbit, however, have so far “won out,” favoring light‑touch oversight and warning that even voluntary frameworks might evolve into de facto regulation.⁵²

With the signing shelved and no clear timeline for revisions, both the administration and the AI industry are now “scrambling to figure out what’s next” for government access to top models and broader AI safety policy.⁸

---

1. Axios – “Scoop: Trump AI executive order seeks early government access to frontier models” – The White House draft EO sought a voluntary framework for early government access to frontier AI models and stronger cybersecurity for critical infrastructure. https://www.axios.com/2026/05/20/ai-trump-executive-order-white-house-infighting

2. Business Insider – “Read the AI executive order Trump didn't sign” – POLITICO’s obtained draft showed voluntary pre‑release reviews of advanced AI models, explicitly disavowing mandatory licensing or preclearance. https://www.businessinsider.com/trump-ai-oversight-executive-order-draft-2026-5

3. Financial Times – “Preventing a ‘Chernobyl moment’ in AI” – Commentary arguing that a White House order mandating testing of frontier AI models would be a significant first step to avert catastrophic failures. https://www.ft.com/content/fe3a33e1-534a-4968-8af3-f768e68874f8

4. Ars Technica – “Trump abruptly cancels EO signing event after top AI firm CEOs declined to go” – Report that Trump canceled the signing after some CEOs declined to attend and that industry feared safety testing could delay or alter model launches. https://arstechnica.com/tech-policy/2026/05/trump-canceled-ai-safety-testing-eo-after-snub-from-tech-ceos/

5. Axios – “Why Trump's AI executive order was pulled” – Behind‑the‑scenes account describing Trump’s aversion to regulation, adviser David Sacks’ opposition, and clashes between accelerationists and AI safety advocates. https://www.axios.com/2026/05/21/trump-ai-executive-order-postponed-why

6. TechCrunch – “Trump Delays AI Security Executive Order: 'I Don't Want to Get in the Way of That Leading'” – Coverage of Trump’s stated concern that the EO’s language might be a “blocker” to U.S. AI leadership and job creation, and disputes over 14‑ to 90‑day model‑sharing requirements. https://techcrunch.com/2026/05/21/trump-delays-ai-security-executive-order-i-dont-want-to-get-in-the-way-of-that-leading/

7. The Verge – “Trump delayed signing AI executive order because he “didn’t like certain aspects of it.” – Trump linked the delay to fears the order could obstruct the “tremendous good” of AI and America’s lead over China. https://www.theverge.com/policy/935694/trump-delayed-signing-ai-executive-order-because-he-didnt-like-certain-aspects-of-it

8. Axios – “Read the AI executive order thwarted by Trump tech allies” – Analysis noting that the industry and administration are scrambling to determine next steps as key questions about government access to top models and AI safety remain unresolved. https://www.axios.com/2026/05/22/ai-executive-order-cancelled-white-house