The legality of training frontier AI in the UK is an open question
British AI founders are building models against a copyright and data regime whose application to training has not been settled. The detail of that exposure is not yet established.
Whether a British company can lawfully train a frontier AI model on data gathered in the UK is being asked as a serious legal question, not a theoretical one.
The European startup publication Sifted has put that question directly: is it illegal to train frontier AI in the UK?
Beyond the question itself, no detail is established here. No statute, exception, case or regulator position can be set out on the strength of what is in hand.
That matters because the answer determines the cost base of every British AI company that trains rather than licenses. If copyright law does not permit training on scraped material at commercial scale, licensing is the only route, and licensing has a price.
Founders in this position should assume the question is live and take their own legal advice before the next training run, rather than after it.