UK lawmakers are warning that Britain’s AI strategy may depend on something it does not fully control: continued access to US-controlled frontier models.
The concern turns AI sovereignty into a practical risk for UK and EMEA enterprises. If critical workflows depend on foreign-controlled models, technology leaders need to know what happens if access terms, permitted uses, or availability change.
The House of Commons Science, Innovation and Technology Committee raised the issue in its July 7 report, Science Diplomacy: Sovereignty, Strategy, and the Global Race. The report cited US restrictions on Anthropic’s latest AI models, and later limits affecting access to OpenAI’s latest models, as evidence that the UK “may not be able to count on its allies” for technologies tied to economic growth and national security.
The warning comes after the government announced a £1.1 billion AI Hardware Plan covering chips, compute, procurement, skills, and investment. That plan may strengthen domestic infrastructure, but recent scrutiny of AI data center readiness shows why compute strategy still depends on power, planning, and capacity. The committee’s concern is about the model layer: compute capacity does not guarantee continued access to frontier models.
Model access becomes the sovereignty test
The committee said the government has “no coherent strategic framework” for using science and technology partnerships to support diplomatic and economic goals. The report also said the UK has not clearly identified priority partners, target technologies, or intended outcomes in critical areas such as AI, quantum, and space.
International partnerships can expand the UK’s access to advanced systems, but they do not guarantee control. Dependence on foreign-controlled systems creates risk when another country can change export rules, security terms, or access permissions.
Dame Chi Onwurah, chair of the committee, said the UK needs “a realistic plan to achieve sovereign capabilities in critical areas” or it could see access “cut off at the whim of its partners,” according to the committee’s July 7 announcement.
The committee is not arguing that the UK should build every part of the AI stack domestically. Its broader point is that the government needs to define where the country must own capability, where it can collaborate, and where partner access is enough.
Enterprise AI risk moves beyond compute
The model-access issue has immediate procurement and resilience implications. UK and EMEA organizations using US frontier models in production should identify which workflows depend on a small group of providers, whether alternative AI models could support priority use cases, and how long migration would take if access narrowed.
That audit should include vendor-contract terms, export-control language, access-termination provisions, data portability, and fallback model options. Business continuity plans should also account for model-level disruption, not only cloud or infrastructure outages.
The government has already moved on domestic AI capacity. Its separate Sovereign AI effort is a £500 million push to back homegrown AI founders with funding, supercomputing access, visa support, and research and development help.
Those investments address pieces of the domestic ecosystem, but they do not settle the model-access question, especially as frontier AI developers keep signing larger AI infrastructure deals to secure scarce compute capacity.
Until that framework appears, model dependence should be treated as part of AI risk management. If a key model becomes unavailable, organizations need a credible way to keep operating.
Read more: As AI agents move deeper into enterprise workflows, the first known agentic ransomware attack shows why model-level risk is becoming a security issue, too.