White House has asked OpenAI and Anthropic not to share their new AI models with the UK testing agency until the models have gone through testing with the US government first, according to a Politico reporter

This fits an emerging pattern of governments asserting jurisdiction over frontier AI model releases before they circulate to foreign allies, echoing how strategic technologies have historically been managed as national security assets rather than commercial products. The operative distinction is between closed-weight model providers, who can gate access and sequence disclosure to regulators, and open-source releases, which are far harder to police; the sector has been bracing for exactly this kind of scrutiny. The counterparty that matters here is the UK AI testing outfit, whose value depends on early access to frontier models; being pushed behind Washington in the queue dilutes its relevance and underscores who sets the terms of the AI safety agenda. Watch for whether this becomes a formal requirement or remains ad hoc guidance, whether the affected labs push back, and any signs of friction with London over the optics of US-first access. As with earlier US-China chip and software controls, compliance burden and approval sequencing are the channels that matter for timelines, while Microsoft and other hyperscalers face second-order exposure through their partnership and infrastructure commitments. The next step is any written guidance from the White House and the response of the affected labs.