AI generated works – who owns copyright?

Having in part I [Artificial Intelligence (AI) – ethical and legal considerations - Saunders Law] already discussed the complex issue of infringement of copyright by AI tech companies and possible defences to such claims, part II will now address the question whether copyright even exists in a work that is created by AI systems and if it does, who may own copyright.

Does copyright exist in an AI generated work?

In general, any work (artistic, dramatic, literary, musical) that may be protected by copyright requires a human author. Works generated by AI are ‘computer-generated’ and therefore do not have a human author. According to section 178 of the Copyright Design and Patents Act 1988 (‘the Act’) ‘computer-generated’ means ‘that the work is generated by computer in circumstances such that there is no human author of the work’. Nevertheless, the Act contains a provision at section 9(3) which states that the author of a computer-generated works shall be ‘the person by whom the arrangements necessary for the creation of the work are undertaken’.

This raises an important question:
Who are the ‘persons making the arrangements’ in relation to works generated by AI technology?  It has been suggested that these persons include the developers who created the machine-learning software, the user who input the prompt to generate the work, or a third party who owns copyright in the prompt.

In the absence of clarification by the courts (or the legislative) in the UK as to who exactly may own copyright in AI generated works, it would be safest for AI tech companies to ask their software developers to assign any rights in the software they are developing to the company and then state clearly in their terms and conditions of use that any content generated by the software can be used by the user/subscriber to the software without restriction.

As noted in Part I AI tech companies cannot be sure whether they own copyright in the output unless they have cleared rights in the dataset they have used to train their AI system.

Furthermore, a prompt can be deemed a substantial part of a copyright work. For example, the CJEU ruled in Infopac I that an extract of a newspaper article comprising 11 words could infringe the original article[1]. I have already discussed in Part I, that copyright in the output generated by the AI system following the prompt may be owned by the copyright owner of the original article because the generated content may be a copy if the AI system has taken a substantial part of the article or alternatively, may be an adaptation of the article

If the ‘output’ is a sound recording, who has rights in the sound recording?

There is currently much discussion in the media about new songs using the voice of well-known musicians. While the voice of musicians is not necessarily protected by copyright, musicians own rights in their recorded performances. Performance rights are infringed if the AI generated sound recording uses a substantial part of a recording of a performance by the artist.

Where new songs do not utilise already existing sound recordings to create an entirely new song, copyright infringement based on rights in sound recordings and artist’s performances are unlikely to be infringed. However, if the artist’s voice is highly recognisable, artists may have a claim for passing off (or trade mark infringement if the artist’s voice is protected as a sound mark) under English law. Passing off provides protection where the public may believe that due to the voice being highly recognisable by the public of that of a popular recording artist, the song may have been recorded by that artist resulting in likely damages to the artist’s reputation or loss of earnings.

It is even easier to enforce a registered trade mark if the infringing voice is identical or similar to the voice registered as a sound mark – for this to be the case, the registered sound mark must be sufficiently distinct from other artist’s voices.[2]

A note of caution to users of AI generated content/AI systems

Unless the provider of an AI software application can guarantee that all IP rights are cleared in the dataset they have used to train their software (currently very unlikely), that they have not breached database rights, privacy rights or confidentiality it will be necessary for users who use AI generated content, for example, to promote their business, or to develop works based on AI generated content or to integrate AI systems in their services, to check very carefully, whether such works infringe third party rights.

It is of course possible to use AI generated content as a starting point for any project to create a piece of new work, as long as the new work is an original creation and does not take a substantial part of an already existing work or is an adaptation of an already existing work protected by copyright.

As to creating new songs using the voices of well-known artists, this is likely to result in a claim for passing off. It is safer to use voices of not so well-known artists but this may not sell records. While the latter may not breach the law, such conduct may damage a business’s reputation. Instead, consider engaging the relevant artists if you like their voices and give them an opportunity to record a new song.

Ultimately, it will be a matter of due diligence and risk assessment to use AI generated content or AI systems for projects or services which are created for commercial purposes. The higher the potential value of your new project or services the greater the importance to minimise the risk. This will include engaging with relevant professionals such as IP lawyers to advise you.

Silvia Baumgart

May 2024

[1] Infopac International A/S v Danske Dagblades Forening [ECLI:EU:C 2009: 465] (Case C-5.08) (Infopac I)

[2] Sound marks are a particular form of trade mark. There are relatively few sound marks registered at the UK IPO as it is difficult to meet the requirement of distinctiveness for the sound mark to be registered. It is likely that the examiner will require evidence that the sound has acquired distinctiveness through use and a substantial part of the public will recognize the sound as the unique voice of a particular artist.


    How can we help?

    Please fill in the form and we’ll get back to you as soon as we can

    We have partnered with Law Share from JMW Solicitors LLP to refer instructions and clients to them, when we are unable to act. By answering yes to this question, you agree that we may pass your details on to Law Share in such circumstances. You are under no obligation to instruct JMW Solicitors LLP after being referred. We may receive a payment from JMW Solicitors LLP further to this referral.