On 18 March 2026, the UK government published its Report on Copyright and Artificial Intelligence, along with an accompanying economic impact assessment. It's the outcome of a process that began back in December 2024, when the government opened a consultation asking whether AI developers should be allowed to train models on copyrighted material under a broad exception, with rights holders required to actively opt out.
That original proposal did not survive contact with the creative industries. Of the more than 11,500 responses received, only a small fraction backed the government's preferred opt-out model, while the overwhelming majority — reportedly around 95% — supported keeping or strengthening existing copyright protection and licensing training data properly. UK Music, along with artists across the industry, campaigned hard against the opt-out approach throughout early 2025.
The upshot in the March 2026 report is, broadly, a pause rather than a resolution. Rather than pushing ahead with the broad text-and-data-mining exception it originally floated, the government says it will keep gathering evidence — commissioning further research into the impact of copyright reform on both the creative industries and the AI sector, and continuing to engage with developers and rights holders on how training data is actually sourced and used in practice.
For engineers and producers, the most useful takeaway is that nothing has been finalised, but the direction of travel favours licensing over a blanket opt-out. That matters directly for anyone selling beats, stems or sample packs: how those files are licensed, and what rights a buyer actually receives, is likely to stay a live commercial question rather than settle into fixed rules any time soon.
There's also a case worth watching in the background. A UK copyright dispute between Getty Images and an AI image-generation company reached a first judgment in late 2025, largely on secondary infringement grounds since the underlying model training took place outside the UK. Permission to appeal was granted in December 2025 on a point of law the court itself called novel and important, with a hearing expected sometime in 2026. However that appeal lands, it's likely to shape how UK courts treat AI models trained overseas but distributed here — which is exactly the kind of question that will eventually determine how music and audio rights are protected against AI training, not just images.
Our approach in the meantime hasn't changed: every beat licence we sell spells out exactly what a buyer can and can't do with the file, and we keep detailed records of everything we produce in-house. Until the law catches up fully, clear paperwork is still the best protection either side has.