Trump signs order for voluntary AI cyber testing

TL;DR:

• Donald Trump has signed an executive order creating a voluntary framework for federal agencies to vet powerful AI models up to 30 days before release.
• The order explicitly rules out any mandatory licensing or preclearance, after Silicon Valley figures lobbied against stricter earlier drafts.
• It marks a partial shift from Trump’s deregulatory stance, driven by cybersecurity fears over models such as Anthropic’s Mythos.

The order asks developers to share their most capable models with government for review, with agencies including Treasury, Defense, Commerce and Homeland Security tasked with securing those agreements. Yet its language insists the US “refuse[s] to stifle this innovation with overly burdensome regulation”, and it bars the creation of any compulsory permitting regime for frontier models.

A hedged shift

The voluntary design reflects a tug-of-war inside Trump’s coalition. An earlier draft reportedly sought models 90 days ahead of release; Trump postponed the May signing after tech leaders including Elon Musk and Mark Zuckerberg pressed him in private calls, saying he did not want to jeopardise the US lead over China. Reuters reported that Treasury Secretary Scott Bessent will coordinate with banks and AI firms to scan software for vulnerabilities, while OpenAI’s Sam Altman said the order “gets the balance right”.

Voluntary federal testing is not new — OpenAI and Anthropic already submit models to the Commerce Department’s Center for AI Standards and Innovation — but the order formalises a more active monitoring posture.

The contrast with the UK is instructive. Britain’s AI Security Institute has pursued an evaluation-led model with in-house technical capacity, though former staff have warned of a widening gap between AI capability and oversight. Where the UK debates how to resource scrutiny, Washington’s question is whether scrutiny should be obligatory at all.

Looking forward

A voluntary regime leaves enforcement resting on goodwill and competitive optics rather than law. For UK readers, it sharpens a strategic divergence: the EU legislates, the UK evaluates through guidance and institutions, and the US is now betting that light-touch, innovation-first oversight keeps it ahead — a wager that looks fragile if a frontier model is misused before anyone volunteers it for testing.