Innovation or infringement: Copyright considerations in the era of AI-generated music

With the rise of Artificial Intelligence (‘AI’), we have entered a new era of music creation. Music that would have taken hours can now be composed in minutes using AI. Amidst this technological evolution, some fundamental legal questions emerge: Is that work capable of being protected by copyright and who does any copyright belong to? Does it infringe existing copyright? How will this impact the music industry?  

1. How can AI tools be used to create music?

AI models are trained on large datasets of existing music data (e.g. songs, melodies, rhythms, beats, chord progressions), analyse those patterns and structures, and then generate new music based on that analysis. Whilst these tools can autonomously create, they are also often used by human composers and musicians to augment their own skills.  

2. Can AI-generated music be protected by copyright?

Original music is protected by copyright - a property right giving the owner (the composer and/or writer, or the person to whom the composer and/or writer has assigned it) an exclusive right to authorise or prevent others from using their work. The Copyright, Designs and Patents Act 1988 (‘CDPA’) is the key UK copyright legislative framework enshrining those rights.

If copyright does not subsist, the work can be copied without the risk of infringement.  In the UK, copyright protection arises from the moment a protectable work is created. 

Originality

Under the CDPA, for copyright to subsist, music must be original. The question therefore becomes - is AI generated music capable of being original? 

The threshold for originality is whether the work is the ‘author’s own intellectual creation’, as re-defined by the European Court of Justice (CJEU) in the 2009 Infopaq case. ‘Intellectual creation’ implies that there needs to be an element of human input for the originality threshold to be met. The CJEU required that the author ‘expresses’ their ‘creative ability in an original manner by making free and creative choices’, stamping their own ‘personal touch’. Whilst there is of course, a ‘human’ element to how these AI tools create music, it is not clear that the human’s conception of an idea or interaction with the tools will be sufficient to meet the ‘expression of creative ability’ standard. Each AI tool functions differently from the other, requiring varying degrees of human input. In light of this, each piece should be looked at individually, with an assessment of the level of human input and its importance to the final piece, before deciding whether it meets the ‘intellectual creation standard’ for originality. If a tool simply generates lyrics from a text prompt (e.g. “create an uplifting chorus for a song about chasing dreams”), that text prompt may be unlikely to be sufficient to meet the standard for originality. 

Interestingly, under the current legislation, even where there is no human author and work is generated by a computer, 'computer-generated work’  is capable of being protected under s.178 of CDPA.  When this was proposed in 1987 Lord Young of Graffham, when debating the Bill, said this is ‘the first copyright legislation anywhere in the world which attempts to deal specifically with the advent of artificial intelligence’. Clearly the ‘computer-generated’ work distinction was intended to apply to works of AI. But there seems to be a contradiction here – how can ‘computer-generated’ work produce an ‘intellectual creation’ where there is no human author? In its 2022 consultation on AI and Intellectual Property, the Intellectual Property Office (IPO) suggested there were no plans to update the law on this, as ‘AI is still in its early stages’. Given the continual rate of development, the Government should act sooner rather than later to properly legislate in these areas.  

Ownership

Determining the owner of the work is tricky – is it the creator of the AI tool, the individual or business which is using the tool, or, if separate, the individual user inputting the prompt? They are all arguably people ‘by whom the arrangements necessary for the creation of the work are undertaken’, and are eligible authors – the CDPA is yet to be adapted to address this issue. In the US, where copyright must be registered by way of submission to the Library of Congress to subsist, the US Copyright Office have refused to protect AI generated works without a human author.  

3. Can AI generated music be capable of infringing copyright?  

What happens when AI creates music that infringes on copyrighted works when it has been ‘inspired’ by existing works? Is AI to be held to a higher standard than human artists in that way? 
 
Last year, an AI-generated track ‘Heart On My Sleeve’, uploaded by a TikTok user, went viral  – resembling the voices of Drake and The Weeknd. In fact, it was created by an anonymous creator ‘Ghostwriter’ using generative AI technology to conjure the voices of familiar pop stars. Universal Music Group (‘UMG’), the artists’ label, was able to remove the content, citing Intellectual Property (‘IP’) concerns. Under UK copyright law, whilst a ‘voice’ itself isn’t strictly protectable by copyright, arguably deep-fake vocal synthesisers used without the voice owner’s consent could constitute passing off (an IP right used to enforce unregistered IP rights), though the application of this is yet to be seen. 
 
Some AI models are trained using artists’ copyrighted content from unlicensed sources, without obtaining the required consent or paying compensation to the rightsholders who own or produce the content. To avoid infringement, AI developers could:

• Purchase the music / obtain a licence from the copyright holder (However, there is a risk that the rightsholder fails to give permission);
• Take a risk in not seeking a licence with the intention to rely upon the defence of sufficient acknowledgment by giving the same to the rightsholder; or
• Use music that has fallen out of copyright protection to train their models, given that copyright protection expires 70 years after the composer/writer’s death. 

Even if an AI model has been trained on copyrighted content for which the AI model creator did have permission, questions arise as to whether the licence which a label or artist provides for their music to be listened to or downloaded actually extends to a licence for it to be used for training an AI model. 

In circumstances where the AI model may be used specifically to mimic an artist’s voice or style in an attempt to pass off new content as having originated from that artist, it is a fairly safe assumption that no artist would have agreed to a licence which allowed access to their music for that purpose. Following the Ghostwriter debacle, UMG released a statement that they consider that they have a ‘moral and commercial responsibility’ to artists to ‘prevent the unauthorised use of their music and to stop platforms from ingesting content that violates the rights of artists and other creators.’ 
 
Since then, music labels have taken more drastic steps to combat unauthorised use of their artists work. Last month, UMG (home to some of the biggest music stars including Taylor Swift and Adele) announced in their Open Letter that it will stop licensing its artists’ music to TikTok due to their artists’ payments being just a ‘fraction’ of what other major social media platforms pay. An interesting factor in their decision, was the growing use of AI-generated sounds on the platform, which ‘dilute’ the royalty pool for human artists. 
 
In the latest battle between Big Tech and the music industry, UMG, and other labels AKKCO and Concord, have filed a copyright infringement lawsuit in the US against AI start-up Anthropic, who created ‘Claude’ a chatbot rivalling ChatGPT, which was allegedly trained to repurpose copyrighted lyrics. Anthropic, in response, are asking for the court to reject the music publishers’ request for an injunction, citing that their AI models are ‘trained using data publicly available on the internet’, and their use of lyrics amounted to ‘fair-use’ of the copyrighted material, which means there has been no infringement. 

‘Fair-use’, termed ‘fair dealing’ in the UK, is a defence to copyrighted material being  used provided that the use is considered fair, reasonable and falls within the permissible purposes under the CDPA. The only specified purpose that AI developers could argue applies is ‘text and data mining for non-commercial research’. Developers are ‘data-mining’ publicly available, albeit copyrighted, music to analyse that data for patterns, to train AI models. However, given that this ‘defence’ only applies to ‘non-commercial’ research, it would be difficult to see how developers could argue that there is no commercial gain from training their AI models. It remains to be seen whether this defence could be successfully applied under UK law.
 
The UK has yet to see any concluded litigation in relation to copyright infringement by AI companies, but it is clear that the UK legislative and regulatory framework is currently ill-equipped to deal with these complex issues. The Government’s report on 'a pro-innovation approach to AI regulation' shows some awareness of the deficiencies. It concluded with a recommendation that ministers from the Department of Science, Innovation and Technology liaise with both the AI and creative sectors to find a solution.

4. The future of AI-generated music

AI- generated music has the potential to disrupt the music industry both in terms of how music is created, distributed and perhaps even protected. Whilst some music purists say that the use of AI is a cheat and somewhat undermines human creativity, musicians have long embraced technology. Autotune was used to manipulate voices (think T-Pain) or digital production tools to sample and to re-purpose music (used as early as 1985 in Kate Bush’s Running Up That Hill). Arguably, generative AI is no different and artists themselves may want to legitimately exploit AI technology.

To ensure that musicians are helped rather than hindered from the use of this technology, the regulations need to be expanded to protect them. If artists’ work is being used in AI models, they should be credited and/or compensated, for instance, using licensing models. There is a delicate balance between allowing AI to develop and protecting the rights of musicians. Whilst the technology continues to develop at an astonishing rate, the fate of AI-generated music will depend on when and how governments and courts respond. 

further information

If you have any questions or concerns about the topics raised in this blog, please contact Melanie Hart or Ben Atkin in our Dispute Resolution team. 

about the authors

Melanie Hart advises on a broad range of corporate, commercial and individual disputes, and represents clients in litigation, arbitration and mediation. Much of her advice involves ‘dispute avoidance’ and developing risk management solutions in order to identify and mitigate risk at an early stage. She is known for giving highly commercial and practical advice, which always focuses on her client’s key objectives, be they commercial, financial or reputational.

Ben Atkin is a senior litigation lawyer and solicitor advocate in the Dispute Resolution team. He advises clients on a wide variety of matters including corporate/commercial disputes, contentious intellectual property, and media related litigation including in relation to defamation, libel and privacy matters, many of which have a high profile in the national press. He has experience in the County Court, High Court and Supreme Court as well as experience in resolving disputes without the need for litigation via various forms of Alternative Dispute Resolution. 

Lavanya Loganathan is a trainee solicitor at Kingsley Napley and recently completed her third seat with the Dispute Resolution team.