Copyright in AI-generated music has been legally ambiguous since the first generative audio tools went to market. Production teams and agencies have been operating in that ambiguity for the past two years, mostly by relying on contractual indemnification from their generative audio vendor rather than on any clear statutory framework. That situation has started to change — slowly, and differently on each side of the Channel.
This post covers the regulatory and legal developments in the UK and EU that are most relevant to commercial production teams using generative audio. It is not legal advice, and the landscape is still evolving. We've tried to be accurate about what has changed and honest about what remains unresolved.
The UK position: computer-generated works and the CDPA
The UK has had a statutory provision for computer-generated works since the Copyright, Designs and Patents Act 1988. Under section 9(3) of the CDPA, the "author" of a computer-generated work is the person who made the necessary arrangements for its creation. This was written for software-generated outputs where there is no human author in the conventional sense — spreadsheet charts, database-generated reports. Whether a piece of AI-generated music qualifies as a "computer-generated work" under the CDPA has been contested since at least 2023.
The Intellectual Property Office's consultation on AI and copyright, which closed in early 2024, produced a response in late 2025 that clarified the government's position: AI-generated outputs can qualify as computer-generated works where there is no human creative authorship in the output itself, but where significant human creative input is present — through detailed prompting, iterative selection, or post-generation editing — the output may attract conventional copyright with the human as author. This is a spectrum, not a binary, and the IPO acknowledged that the line between "computer-generated" and "human-authored with AI assistance" will require case-by-case assessment.
For practical purposes, what this means for a production team commissioning generative audio is: the copyright status of the output depends on the level of human creative involvement in the generation process. A fully automated generation from a brief with no human curation of the output has a different legal status than an iterative process involving multiple rounds of human-directed refinement.
What changed in early 2026 for UK commercial clearance
The more immediately practical development was the update to standard advertising clearance procedures. The ISBA updated its model music licensing template in January 2026 to include an explicit representation clause for AI-generated music — requiring the music supplier to warrant that any AI-generated content does not infringe third-party rights through training data provenance, and that the supplier maintains records of the models used and the training data provenance claims made by those models.
This does not resolve the underlying copyright questions, but it shifts contractual liability clearly to the vendor. For buyers, the practical implication is: if you are commissioning generative audio for broadcast advertising, your contract with the audio vendor should include an explicit AI training data indemnification clause. If your existing vendor agreements predate January 2026 and don't contain such a clause, it's worth updating them.
The EU position: the AI Act and creative works
The EU AI Act, which entered its implementation period in 2024, applies to AI systems rather than to works produced by those systems. Its copyright provisions — specifically Article 53(1)(c) concerning transparency requirements for general-purpose AI models — require providers of general-purpose AI models to make publicly available a sufficiently detailed summary of the content used for training. This is aimed at enabling copyright holders to assess whether their works were used in training without consent.
What's new in early 2026 is the first enforcement actions under this transparency requirement. Several major collecting societies in France and Germany have filed formal complaints against AI music generation providers who have not complied with the training data summary requirement. None of these have resolved yet, but they represent the first attempt to use the AI Act's transparency obligations as a mechanism for copyright enforcement rather than as a standalone compliance requirement.
For production teams operating across EU markets, the relevant question is whether the generative audio tool they're using can demonstrate training data transparency compliance under Article 53(1)(c). If it cannot, the tool may face enforcement action that could affect service availability. This is worth including in vendor due diligence for any EU-facing production work.
Training data provenance: what vendors should be able to tell you
The practical test of a vendor's compliance posture is whether they can answer three questions: What data was the model trained on? How was that data acquired? Do they have licences for the training data, or do they rely on a copyright exception such as text and data mining?
The text and data mining exception in UK copyright law (section 29A CDPA) and the EU's similar exception under the DSM Directive apply to training conducted for research purposes. Whether they apply to commercial AI model training is not settled law in either jurisdiction. Vendors who rely on this exception as their training data justification are carrying legal risk that may eventually transfer to their customers through indemnification gaps.
We train on licensed music catalogues and original commissioned content with explicit licensing for AI training use. This is not the cheapest training data approach, but it is the one that produces a defensible indemnification position. We can answer all three of the above questions clearly, and we include that documentation in enterprise customer agreements.
What remains unresolved and what to watch for
Three significant questions remain open in both jurisdictions. First: whether AI-generated output that closely resembles a copyrighted work in style constitutes infringement, even where no specific audio content has been copied. Style is not protectable under current copyright doctrine, but the line between style imitation and reproduction of protected expression is contested. Second: how the moral rights framework in EU member states applies to AI-generated works — specifically, whether a human who significantly shaped an AI output through prompting and curation has moral rights in the result. Third: cross-border enforcement, where a UK studio uses an EU-based model to generate music for a US campaign — which jurisdiction's copyright law governs.
None of these will resolve quickly. The safest operational position in the interim is: use vendors with documented training data licences, ensure your contracts include explicit AI indemnification clauses, and keep records of the generation parameters and human editorial decisions that went into any commercially deployed AI music. That record-keeping may matter if a copyright challenge arises after the fact.
