For a UK writer, what is and is not allowed depends on three sets of rules – British, American and European – and all three are unsettled.
At the end of 2024 the UK government asked how copyright law should deal with AI training. Its favoured answer was a new exception letting AI firms mine copyrighted work unless the owner actively opted out. The creative industries pushed back hard, and in early 2026 ministers admitted that favouring that option had been a mistake. The government’s March 2026 report dropped it and left the existing law in place while it waits to see how the lawsuits and overseas rules develop. Nothing on AI and copyright made it into the 2026 King’s Speech, so for now the law is unchanged, though the government has been careful not to rule out returning to the question.
The US has no AI-specific copyright law, so the fights are happening in the courts. The early rulings went the AI companies’ way on training, but they turned heavily on whether the use harmed the market for the original work, and the US Copyright Office has said many uses, music among them, may not be fair use. With more than a hundred cases running, the American position will be built up slowly, one case at a time.
The EU’s main tool is the AI Act. It makes the big AI model providers publish a summary of what they trained on and obey EU copyright law, including a creator’s right to refuse text and data mining. From August 2026 the EU can enforce this with fines of up to 3% of a company’s worldwide turnover. In March 2026 the European Parliament also passed a resolution calling for AI to be developed without trampling copyright or the people who depend on it.
None of this is finished. What the three share is pressure on AI companies to disclose what they used and to pay for it; how far that pressure goes, and how soon, is still being worked out.
