In this article Michael Bywell, independent consultant and Partner at New Media LLP summarises the direction of travel on AI and music copyright issues.
Introduction
The questions here concern whether recordings, lyrics and compositions can be copied and processed to train generative AI models and then exploited without the permission of or payments to rightsholders. These issues are being tested in cases against music AI platforms in several jurisdictions including the US, Germany, Canada and Denmark. And in the UK, the government concluded a consultation earlier this year on how UK copyright law should respond to generative AI which is discussed below. The Getty case has also made waves and has relevance even though it is not a music industry case.
The scale of the issue
The actual scale is uncertain mainly because the training statistics are not publicly available nor readily disclosed. However, to give some idea, Suno admits to having trained its model on ‘tens of millions’ of audio files and, on usage, claims that more than 100m people have made music using Suno since it started up in 2022.
US
In the US, Universal, Warner and Sony (the ‘Big 3’ music companies) have all sued Suno and Udio. Suno and Udio are GenAI platforms that allow anyone to create full length music tracks from simple prompts. Some cases have settled and new licensing arrangements have started to appear. Universal and Warner (but not Sony) have settled with Udio; and Warner (but not Universal or Sony) has settled with Suno. We are also seeing artists bringing claims asking for more money of the back of these cases and deals.
Germany
In July 2026 a German court ruled against Suno in relation to the unauthorised use of six well known songs including two Boney M tracks and the song ‘Mambo No. 5.’ The case was brought by GEMA, the German collecting society on behalf of composers, lyricists, and producers. The Munich Regional Court found evidence of copying without permission at the training stage; memorisation inside the model meaning reproductions were stored on the model; and evidence of further infringements at the output stage. In relation to the latter, these infringements were supported by tests that GEMA had undertaken during which they entered open-ended prompts with song title, lyrics, and description of style, and received back recognisable musical elements of original works. Suno argued, unsuccessfully, that the model was learning mathematical patterns and relationships only but the court said that a recognisable form of a work could be generated so there was evidence of reproduction for copyright purposes. This was a robust and bold decision by the Munich court not least because it took it upon itself to consider both German and US copyright law. The US element was relevant to training that took place on US soil. There are reports that Suno is considering an appeal from that decision.
UK
The Getty Images v Stability AI case [2025] EWHC 2863 (Ch) is worth noting even though it is not a music case. It concerns alleged infringement by Stability AI of Getty’s rights over photos, videos, illustrations and digital artwork. The primary infringement claim (data scraping and copying to train the model) was dropped because there was no evidence that training had taken place in the UK. The secondary infringement claim relating to the output or commercialisation of scraped and copied data failed because there was no evidence that images were stored on the model. Technically speaking, the court held that the model’s numerical weights (numerical values representing patterns learned during training) were not themselves copies of the Getty images. So there was no infringement. There is an appeal which is due to be heard next month.
UK government consultation
On 18 March 2026 the UK government published its “Report on Copyright and Artificial Intelligence”. In short, a stocktake of how UK copyright law should respond to generative AI:
Copyright exception?
The report says there will be no copyright exception to allow training on models without the permission of rights holders. Originally the UK government had said there would be an exception coupled with an opt-out right for rights holders. However there was considerable push back with the creative community saying that opt out would not work and in truth meant ‘permission by default.’ It would be too hard for copyright owners (e.g., illustrators, photographers, musicians or writers) to monitor potential infringements and identify (i) that their work had been scraped, (ii) every GenAI developer using it, and (iii) register objections. Furthermore, that GenAI companies should be required to seek permission in the first instance and it was self-evidently wrong to allow those companies to utilise the content of others without payment and then use it to compete against those same rights holders. The government listened to those concerns and changed its mind.
Transparency on what is being used to train models
Reflecting the objections already highlighted above, the report concludes that there is a need for more transparency on what material is being used to train systems. Artists and rights holders support this because it makes it easier for them to discover whether their work has been used and enables them to object to unauthorised use, seek payment and negotiate licences where possible. The government is therefore looking at best practice for input transparency. Legislation has not been ruled out.
Licensing
The report recognises licensing as a potential mechanism for ensuring rights holders are paid when their work is used in GenAI models. The government does not propose intervening and instead will see how the market develops.
Digital replicas
This concerns AI’s ability to replicate someone's voice, face, likeness and identity as a performer - for example, where GenAI generates a completely new song using a synthetic voice that is recognisably that of a human artist. There may be no copying of an artist’s actual recordings or performances but one or more of the artist’s characteristics may have been used. For a singer, this could be very damaging because it is unauthorised use of their identity and could give rise to reputational and financial damage. It might be a terrible or politically motivated song that the original artist does not want to be associated with. The rogue track might also compete with the true artist and potentially deprive them of income.
The report recognises that UK law does not currently cover every situation where someone's voice or likeness is replicated without consent and because of this the government proposes exploring whether the UK should introduce a new “digital replica” or “personality” right.
Why are we not seeing a tidal wave of cases?
Given the amount of music said to have been scraped in recent years and the potentially adverse consequences for artists why is it that we are not seeing more litigation? Yes there are some notable cases but not a lot in the grand scheme of things. There are a few probable reasons for this:
First, the challenges involved in even knowing what has gone into training sets and which outputs have derived from copyrighted material. Without transparency and disclosure obligations how are labels, artists and others supposed to know that their rights have been infringed and how do they get access to platforms to interrogate them? This seems like the main blocker.
Secondly, rights holders may be adopting a ‘wait and see’ attitude and monitoring current legal cases, legislative developments, the enforcement of existing laws and the evolution of licensing deals.
Thirdly, commercially speaking, the big labels and artists may want their catalogues to be used provided that exploitation is accompanied by more licensing fees. They may not want to extinguish or kill off the use of their material altogether.
Finally, smaller independent rights holders may not have the financial ability to investigate and bring claims even if they could somehow get access to platforms to conduct tests.
Future legal claims
In England and Wales the basic statutory period for bringing a copyright claim is six years. So depending on how case law develops over the next few years there will still be a good amount of time for potential claimants to recoup losses going back in time unless, for example, those losses are wrapped up in new licensing deals.
Crystal ball gazing
There are people far more expert than I who may be able to predict what will happen in future. My best guess is that those with commercial leverage and financial muscle will get close to a win:win situation for both sides – platforms get the music to exploit and the rights holders get even more value from their catalogues via licensing fees as more and more people utilise music platforms. Hopefully smaller independents will be able to tuck into the ‘slipstream’ of these larger players and deals in order to receive their fair share. And above all perhaps, let’s hope that originality and creativity continues to flourish and is not deterred as the legal, technical and commercial aspects of all of this get straightened out.
Thanks for reading.
Michael Bywell, October 2026