George Williamson, the new head of the Alan Turing Institute, has delivered the clearest articulation yet of Britain's AI dependency problem. The former head of His Majesty's Government Communications Centre — the secretive body that builds gadgets for UK spy agencies — told the Guardian that relying on foreign AI systems that "could be restricted or even switched off" is an untenable position for a sovereign nation. His argument is straightforward: if AI is now embedded in defence, healthcare, and critical infrastructure, then depending on systems you cannot inspect, control, or guarantee access to is a structural vulnerability, not a policy choice. The trigger events are real and recent. In June, Washington temporarily blocked Anthropic's latest models from release outside the US. The US has reportedly asked leading AI labs to withhold models from UK AI Safety Institute testing until American counterparts have first look. These are not hypothetical risks — they are access controls already being exercised by an ally. Williamson's framing: "We've seen now that both political and commercial pressures mean we may not have access in all the circumstances when we want it." The strategic pivot at ATI reflects a government-imposed direction change. In 2025, then-technology secretary Peter Kyle wrote to ATI's chair demanding the institute refocus on defence and security. Williamson's predecessor Jean Innes resigned soon afterwards. Williamson is threading a needle — accepting the national resilience mandate while arguing it encompasses climate and environment alongside defence. "For me climate is as much a national security issue as defence," he said, explicitly keeping the door open for work the Starmer government wanted sidelined. On the existential risk debate, Williamson was pointed. Anthropic recently predicted AI had a greater than 10% chance of wiping out humanity by 2030. Williamson's response: "I'm not sure it's helpful trying to assign particular percentages to it because I have not seen anyone show me a convincing evidence of how they've arrived at the number they then start throwing around." This is a former intelligence official publicly questioning the epistemic basis of the AI safety community's most alarming claims — notable given ATI's position as a potential arbiter of such risks. The China dimension adds a second axis of concern. Williamson cited Chinese AI models that censor historical events as evidence that values are baked into systems. "Try asking some of these models about some events in the recent past in China and you're not going to get an answer that's particularly helpful if you're doing GCSE history." The argument extends beyond censorship to the structural question of whose worldview AI systems encode and reproduce at scale. Williamson's proposed solution is deliberately modest: not frontier model competition but "nationally owned IP" and smaller models available to UK startups. "We can be ambitious here without trying to copy what frontier labs are doing." This echoes Christine Lagarde and Ursula von der Leyen's recent calls for European AI self-sufficiency, positioning Britain alongside a broader Western reckoning with US-China AI dominance. The question is whether "smaller models for national use" constitutes genuine sovereignty or a consolation prize that leaves the UK permanently two generations behind the frontier. The tension at the heart of this story is unresolved: ATI is a registered charity funded primarily by government, now tasked with building national AI resilience while explicitly acknowledging it cannot match frontier lab resources. Williamson frames ATI as an "objective and helpful" arbiter and partner to government. Whether a government-funded institute can be simultaneously objective and strategically aligned remains the oldest question in science policy, and no speech resolves it.
