Back to stories
Policy

Trump Postpones AI Security Order That Would Force 14–90 Day Pre-Release Model Review

Michael Ouroumis2 min read
Trump Postpones AI Security Order That Would Force 14–90 Day Pre-Release Model Review

President Trump on Thursday abruptly postponed the signing of an AI security executive order that would have required frontier labs to hand advanced models to the federal government 14 to 90 days before public release. The signing was scheduled for Thursday afternoon; Trump told reporters in the Oval Office he scrapped it because "I didn't like certain aspects of it" and worried the language "could have been a blocker."

What the order would have mandated

The draft framework went well beyond today's voluntary posture. Per reporting on the language, it would have:

In practice, that converts a hands-off regime into a structured pre-deployment gate — with a review window long enough to slip a major model launch by a full quarter.

Why now: Mythos and GPT-5.5 Cyber

The order was reportedly motivated by recent releases, including Anthropic's Mythos and OpenAI's GPT-5.5 Cyber, both described as capable of rapidly identifying and exploiting security vulnerabilities. The administration had taken a largely hands-off stance on AI until those cyber-capable models sharpened national-security concerns.

The CAISI regime stays in place

Even without the order, the Center for AI Standards and Innovation (CAISI), housed under the Commerce Department, already runs voluntary pre-deployment testing. On May 5, Google DeepMind, Microsoft and xAI signed agreements, joining OpenAI and Anthropic — which signed similar deals roughly two years ago under the Biden-era US AI Safety Institute. Five frontier labs now participate, and CAISI has completed more than 40 assessments, sometimes testing models with safeguards reduced or removed and in classified environments.

Trump's calculus

"We're leading China, we're leading everybody, and I don't want to do anything that's going to get in the way of that leading," Trump said. Reports also flagged a more prosaic reason: too few tech CEOs were available for the signing-ceremony photo op on short notice. The shelved order is separate from the December 2025 executive order targeting state AI laws, which was signed.

What changes for builders

For now, nothing binding — release cadence stays governed by voluntary CAISI agreements, not a mandatory 14-to-90-day clock. But the draft signals where US policy is heading. If revived, a pre-release review window would directly reshape launch timelines, API availability dates, and enterprise procurement planning around frontier models. Teams building on frontier APIs should treat a government review gate as a live scenario for late-2026 roadmaps, even though Thursday's reversal buys the labs more runway.

Learn AI for Free — FreeAcademy.ai

Take "AI Essentials: Understanding AI in 2026" — a free course with certificate to master the skills behind this story.

More in Policy

arXiv Will Ban Authors for a Year if They Submit Unchecked AI-Generated Papers
Policy

arXiv Will Ban Authors for a Year if They Submit Unchecked AI-Generated Papers

The preprint server's computer science chair Thomas Dietterich announced a one-year submission ban for authors who leave telltale LLM artifacts — hallucinated references, stray meta-comments — in their papers.

4 days ago3 min read
Pope Leo XIV Creates Vatican AI Study Group Ahead of First Encyclical
Policy

Pope Leo XIV Creates Vatican AI Study Group Ahead of First Encyclical

Pope Leo XIV established an in-house Vatican study group on artificial intelligence on May 16, signaling that his forthcoming first encyclical — signed 135 years to the day after Leo XIII's Rerum Novarum — will place AI at the center of modern Catholic social teaching.

4 days ago2 min read
Anthropic Paper Warns 2026 Is 'Breakaway' Year for US AI as China Closes In
Policy

Anthropic Paper Warns 2026 Is 'Breakaway' Year for US AI as China Closes In

Anthropic's '2028: Two Scenarios for Global AI Leadership' paper, published May 14, argues the US has 12–24 months of lead time to tighten chip export controls or risk ceding AI norm-setting to Beijing.

6 days ago2 min read