Who actually benefits from an AI licensing regime?

Britain hosts world-class AI researchers and one of the global top AI labs, yet zero general-purpose LLMs have ever been trained on UK soil. This is because commercial text and data mining (TDM) – which every frontier model abroad relies on – is illegal here.

A compulsory-licensing regime forces developers to hunt down and pay the right-holder for every paragraph, image or riff their model scans. This is a significant burden and a reason why general-purpose LLMs have not been trained in the UK. Since copyright does not bind globally, most labs choose to train in jurisdictions that have TDM exceptions and sell their products to the UK to avoid this regulatory burden.

If Parliament tried to block access to non-compliant models, most providers would just geoblock the UK. That would cut the country off from frontier AI, driving talent, capital and tax receipts overseas almost overnight.

Creators would hardly profit from compulsory licensing: even in a perfect system a 10,000-word essay might earn twenty pence, a lifetime of illustrations about £50.

Meanwhile, the upside we forfeit is enormous: mainstream studies put generative-AI diffusion at ~1% extra productivity growth, roughly £28.5 bn a year. A licensing drag of just 30-50 base points wipes out £11.5-19.7 bn annually, compounding to £106-175 bn by 2035.

A one-line fix, deleting "for non-commercial research" from §29A CDPA, would align the UK with Japan, China, and the US, letting any lawfully-accessed work be mined while leaving public reproduction fully protected.