British Prime Minister Keir Starmer personally asked the United States to exempt Anthropic from new AI export controls. That request has been turned down, according to officials familiar with the matter. The decision threatens to slow UK tech innovation and put small firms at a disadvantage.
Why the carveout was sought
Anthropic, an AI safety company based in San Francisco, develops large language models that UK startups and research labs rely on. Starmer’s team argued that blocking access to Anthropic’s technology would hurt British competitiveness in a field where the country hopes to be a global leader. The carveout would have let UK companies keep using Anthropic’s models without the restrictions the US has placed on exports to China and other rivals. But Washington saw no reason to make an exception.
Impact on UK tech
The refusal lands hardest on small and medium-sized enterprises. Big UK firms can sometimes license models through US partners or pay for cloud access from American data centres. For smaller shops, the extra cost and bureaucracy might kill projects before they start. One startup founder, who asked not to be named because talks are ongoing, said the decision “leaves us scrambling for alternatives that don’t exist yet.” The government had been counting on AI to drive the post-Brexit economy; now that bet looks shakier.
Britain’s own export rules are also being tightened. The National Security Act gives ministers power to block technology transfers, and the Cabinet Office has already flagged AI models as a controlled item. But without access to the most advanced US-developed models, UK researchers worry they’ll fall behind in safety testing and fundamental research.
Global tech dynamics
The rejection signals that the US intends to keep a tight grip on cutting-edge AI, even with close allies. Other European countries have been watching closely. France and Germany have both pressed for exemptions for their own AI labs, so far without success. If Washington holds the line, the result could be a fragmented market where only American companies have the latest tools. That might accelerate the development of homegrown alternatives in Europe and Asia, but it will take years and billions of dollars.
For now, the UK’s AI Safety Institute — set up after last year’s Bletchley Park summit — will have to rely on models it can get, not the ones it wants. Whether the institute can still attract top talent without access to frontier systems is an open question. The next meeting between US and UK officials on the export rules is scheduled for early June, but no one expects a reversal.
