White House weighs pre-release vetting for new AI models

TL;DR:

• The White House is considering an executive order that would establish an AI working group of officials and tech executives to review new models before public release, the New York Times reported.
• Concerns about Anthropic’s Mythos model — and its capacity to identify and exploit cyber-security flaws — are reportedly driving the rethink.
• Resultsense view: a US pre-release review regime would draw closer to the structure already operated by the UK’s AI Security Institute, narrowing one of the key transatlantic policy differences UK firms have been navigating since the Trump administration scrapped Biden’s 2023 executive order.

Reuters relayed the New York Times report on Monday: Trump administration officials are weighing whether to introduce formal government oversight of frontier AI models. A White House spokesperson called the discussion of executive orders “speculation” but did not deny the substance. The reported concern is the cyber-attack capability of Anthropic’s new Mythos model, which security experts have said could materially upgrade the speed and sophistication of code-level vulnerability discovery.

A reversal in posture

If implemented, the move would be a striking shift. On taking office in 2025, the president revoked Biden’s October 2023 executive order, which had used the Defense Production Act to require developers of high-risk AI systems to share safety-test results with the federal government before deployment. The administration’s July AI blueprint pushed in the opposite direction — loosening environmental rules, accelerating exports to allies and prioritising US lead over China.

A March policy submitted to Congress urged lawmakers to pre-empt state-level AI rules and protect children from harm-related risks, but did not reintroduce a federal pre-release review. The reported new working group — bringing together tech executives and government officials — would mark the first material step back toward systematic oversight under this administration.

UK and European angles

For UK readers, the structural parallel is the AI Security Institute (AISI). Founded under the Sunak government as the AI Safety Institute and renamed in 2025, AISI operates a pre-deployment evaluation programme for frontier models, with formal access agreements covering most major labs including Anthropic, OpenAI, Google DeepMind and Meta. A US working group of the kind the Times describes would not reach AISI’s depth on day one, but it would create a counterpart that has been notably absent from the post-Biden landscape.

The European Commission, separately, has reportedly been in dialogue with Anthropic about bank testing for Mythos-related vulnerabilities, suggesting that Mythos is now driving multiple regulatory responses concurrently. UK financial-services firms operating across the channel are likely to see overlapping testing requirements emerge as a result.

Looking forward

Whether this reaches a signed executive order — or stays in deliberations — is the immediate question. UK firms running enterprise procurement decisions on US-headquartered frontier models should expect a fresh round of vendor governance disclosures, particularly around cyber-capability declarations, as US, UK and EU regulators converge on a common set of pre-release expectations.