Trump set to sign AI oversight order with 90-day pre-release model review

TL;DR:

  • US President Donald Trump is expected to sign an executive order on AI and cybersecurity as soon as Thursday, creating a voluntary framework for frontier-model developers to share advanced models with the US government up to 90 days before public release.
  • The order would also give pre-public access to critical-infrastructure providers such as banks, and is the outcome of weeks of negotiation between the White House, the National Cyber Director Sean Cairncross, and AI companies including OpenAI, Anthropic and Reflection AI.
  • The framework represents a compromise between MAGA-aligned activists pressing for mandatory government security testing of “potentially dangerous” AI and tech-industry allies including Marc Andreessen and David Sacks, who resisted formal requirements.

The signing — and Reuters’ reporting earlier the same day that the White House Office of the National Cyber Director briefed AI firms in a separate meeting on plans for model review — together set the new regulatory baseline that UK firms have to plan against. The trigger was Anthropic’s Mythos and OpenAI’s GPT-5.5-Cyber, both flagged by their developers as significantly more capable at cyberattack support than predecessors.

A 90-day window changes how UK enterprise buyers should think about new models

For UK firms procuring frontier AI, the practical implication is that any covered US-developed model now passes through a federal review window before launch. UK financial-services customers will likely benefit from the same pre-public access being extended to critical-infrastructure providers, but the framework is US-led: no equivalent UK pre-release window exists, beyond the voluntary AISI testing regime announced in 2024. That creates an interesting asymmetry — the US’s “voluntary” framework may end up giving Goldman Sachs more advance access to Mythos than Lloyds gets, and that is the kind of gap UK regulators and Treasury will need to address quickly.

The political compromise behind the order is also notable. MAGA activists including Steve Bannon and Amy Kremer had pushed for mandatory government security testing; venture capitalist Marc Andreessen and former AI tsar David Sacks resisted. The middle ground — voluntary engagement with a structured 90-day window — pleases neither flank fully and is likely to face renewed pressure after the next model release that demonstrates a capability jump. The Center for AI Standards and Innovation, the Commerce Department body taking the lead, has already received commitments from Google, xAI and Microsoft to submit AI models for security testing — though Reuters noted that those details disappeared from the department’s website without explanation.

Looking forward

UK readers should watch three things: whether the AI Safety Institute formally aligns its voluntary regime with the US 90-day window, whether the Treasury responds with parallel critical-infrastructure access for UK banks, and whether the European Commission frames a competing position under the EU AI Act before Brussels’ summer recess. For UK SMEs buying frontier-model access from US labs, the launch cadence of major models may slow modestly, but the longer effect will be on how confidence-and-resilience claims are documented in supplier contracts.

Share this article