UK Needs Its Own AI to Avoid Foreign Shut-Offs, Turing Chief Warns
George Williamson, head of the Alan Turing Institute, says the United Kingdom must develop its own artificial intelligence rather than depend on foreign systems that could be switched off at short notice. The warning, published on 10 October 2026, presents domestic capability as a matter of national resilience. The Guardian
Williamson said Britain should not depend solely on AI systems it cannot inspect or control, or that could be restricted or switched off. Inspect means being able to audit how a model behaves and how it handles data. He named defence, national security, critical infrastructure and healthcare as the services where he says that dependence carries risk. Control is the issue. Access cannot be assumed.
He framed the argument around large US and Chinese leads in the field. In his account, political and commercial pressures mean London may not have access to AI in all circumstances when it wants it. His prescription is more national agency over access to technology, and a stronger sovereign position.
The Alan Turing Institute is the UK's leading AI research body. It came under new leadership after the UK government, its main source of funding, demanded a strategic shift. Williamson, formerly head of His Majesty's Government Communications Centre, a government body that makes gadgets for UK spy agencies, took over as chief executive with a mandate to reorient the Institute toward resilience and value. The funder had told the Institute to offer more value for money, and in April the Institute's board was reminded of its legal duties by a watchdog.
Two recent transatlantic frictions shape the warning. In June, Washington temporarily blocked the release outside the US of Anthropic's latest models. The US also reportedly asked leading AI labs to withhold their models from testing by the UK's AI Security Institute until they had been examined by its US counterpart. Williamson has set out related thinking in Institute posts including Sovereign AI: building resilience through choice and How do we develop trusted AI for the UK's highest stakes use cases?
The broader context here is a shift from AI policy centred on safety testing and research excellence to AI policy centred on continuity of supply. Inspection matters because high-stakes deployment needs audit of behaviour, failure modes and data handling. Control matters because inference, updates, licences and export permissions sit with vendors and their home governments. Like renting a machine you cannot fix yourself, a system can remain technically in existence yet become operationally unavailable through restriction, throttling, licence change or delayed release.
Looking at what this means for London, the tension is alliance management versus autonomy. The UK procures, evaluates and deploys heavily within a US-led technology ecosystem, and decoupling is neither stated as an objective nor plausible as an outcome. The sovereign position Williamson describes is better read as leverage inside interdependence: domestic models, evaluation capacity and deployment expertise that keep critical functions running if external supply tightens, and that strengthen Britain's negotiating position when standards, testing sequences and release timelines are set elsewhere.
In my view, the test will be procurement rather than rhetoric. Resilience requires systems trusted enough for defence and health use, maintained over time, and integrated into existing operational chains. That implies sustained funding, compute, talent and a customer in government willing to buy and deploy. Without that demand signal, sovereign capability remains research output. With it, the Turing Institute's pivot could reshape how Whitehall defines value for money in AI, from publications and partnerships to assured access under pressure.
