The Gruffalo goes to court: how might UK courts approach the emerging AI copyright dispute?

read time: 8 mins read time: 8 mins
20.11.26 20.11.26

An emerging AI copyright dispute could see the Gruffalo heading a long way from the Deep Dark Woods. Axel Scheffler, illustrator of the iconic children’s book series, has threatened legal proceedings against AI firms concerning AI-generated images of his renowned character. Whilst Scheffler has proposed bringing proceedings in Germany, it begs the question: how would a similar claim fare under UK copyright law?

Through analysis of the arguments raised in Getty v Stability (the current appeal will not effect this analysis), this article considers the key hurdles a claimant may face when seeking to establish copyright infringement against a generative AI model.

Stage one: choosing the defendant

From the outset, the most important question is who should be liable? Does responsibility for potential copyright infringement lie with the AI developer or the end user? Identifying a defendant with sufficient involvement to attract liability may therefore be a significant hurdle in itself.

The obvious reason for claimants to pursue a developer is their availability of assets as they are more likely to have readily available assets for enforcement purposes. Further, given the ever-growing reach of AI platforms, pursuing one developer as opposed to pursing each individual who has accessed the platform to copy protected works is logistically easier.

Alternatively, there is a significant jurisdictional issue with pursuing AI developers who may be based outside of the UK. A claimant may therefore choose to pursue individuals who use an AI platform to produce infringing copies of protected works. However, individual defendants are unlikely to have the same volume of assets to pursue and, as noted, the volume of people with access to AI platforms could make enforcement of any judgment practically impossible.

For the purposes of this article, based on Scheffler’s indication that he would pursue AI firms, we will focus on the same and not consider enforcement against individuals at this stage.

Stage two: copyright subsistence

Next, Scheffler would need to establish that copyright subsists in the work. Sections 3 - 8 deal with the various types of work protected by the Copyright Designs and Patents Act 1988 the piece of legislation states what qualifies for copyright protection.

As the Gruffalo illustrations are "artistic works" within section 4 of the Copyright Designs and Patents Act 1988 , this is unlikely to present difficulty.

Scheffler would also need to establish that the work is his own intellectual creation, proving that he was able to stamp the work with his personal touch, as established in the EU judgments Infopaq and Cofemel and most recently confirmed in THJ Systems v Sheridan. Finally, as Scheffler resides in the UK, and did so when he illustrated the Gruffalo, the qualification requirements for copyright protection would also be met.

Stage three: primary infringement

As in Getty v Stability, the primary infringement analysis can be divided into two issues: copying during training and copying in the output. Although Getty ultimately abandoned its primary infringement claims, the case provides a useful framework for assessing whether an AI model trained on the Gruffalo illustrations could infringe copyright under sections 16 and 17 of the Copyright Designs and Patents Act 1988 .

Copying during training

The first aspect of this argument concerns the scraping of material protected by copyright to create datasets used to train AI models. Getty alleged that millions of images were copied for this purpose. To advance such a claim, Scheffler would need to show that the Gruffalo illustrations were used during training as part of a dataset and that copies were made without his authorisation. Given the opacity of many AI training processes, establishing precisely what material was used and how it was processed may prove evidentially challenging, as was the case in Getty v Stability. AI developers use vast datasets to train their models. Establishing what image was used, when it was used and how it was used during the training is extremely difficult.

A further question is whether infringement occurs during training itself. While Getty argued that works are copied when ingested and processed by an AI model, the court observed that the model retains statistical relationships rather than copies of the underlying works. The key issue would therefore be whether anything amounting to a reproduction of the illustrations remains once training is complete.

Contrastingly, Penguin Random House's recent German claim against OpenAI for unlawful copying of ‘Coconut the Little Dragon’ alleges that AI outputs demonstrate unlawful memorisation of the illustrator’s work. If similar evidence could be produced in relation to the Gruffalo, Scheffler may argue that the model contains more than abstract patterns and instead preserves a sufficiently identifiable expression of his work.

The key question would therefore be what exactly remains of the Gruffalo artwork once training is complete? If the model no longer contains copies of the works, infringement may need to be established by reference to the training process itself rather than the AI system.

Copying in the output

Even where copyright works have been used to train an AI model, it does not necessarily follow that the resulting outputs infringe copyright. In Getty v Stability, a key challenge was proving that the generated images reproduced a substantial part of the copyrighted works, rather than merely reflecting patterns learned during training.

Similarly, Penguin Random House have alleged that the OpenAI outputs were "virtually indistinguishable" from the claimant's works. If established, this would suggest reproduction of protected expression rather than mere inspiration.

For Scheffler, the critical question would be whether the AI-generated output reproduces a substantial part of the original Gruffalo illustrations rather than merely imitating their style.

Jurisdiction issues

IP rights are jurisdictional and therefore where the infringement occurs is key to any claim. In Getty v Stability, Getty ultimately abandoned its primary infringement claims in part due to jurisdictional difficulties, as much of the alleged training activity occurred outside the UK.

Scheffler could face similar challenges if any AI model trained on the Gruffalo illustrations was developed overseas, since primary infringement under the Copyright Designs and Patents Act 1988 requires a relevant infringing act to occur within the UK.

Stage four: secondary infringement

Infringing copy

For the purposes of secondary infringement, under section 27 of the Copyright Designs and Patents Act 1988 an article is an infringing copy if its making constituted an infringement of the copyright in the work in question, including where it was made abroad and would have infringed had it been made in or imported to the UK.

In Getty v Stability, there was debate as to whether an AI model, or outputs distributed online in an intangible form, could constitute an infringing copy for these purposes. Although the court accepted that an infringing copy need not be tangible, it rejected Getty's argument that the AI model itself was an infringing copy.

The same issue would arise in the context of the Gruffalo. Unless the AI model or its outputs can be shown to reproduce a substantial part of the protected illustrations, a secondary infringement claim is unlikely to succeed.

Importing infringing copies

Section 22 of the Copyright Designs and Patents Act 1988 may be relevant where an AI-generated output reproduces a substantial part of the Gruffalo illustrations and is generated or made available overseas before being imported into the UK.

In the AI context, an interesting question is whether an AI-generated image that reproduces a substantial part of a copyright work could constitute an infringing copy for the purposes of section 22. If an AI system generated outputs reproducing a substantial part of the Gruffalo and those outputs were brought into the UK in the course of business, section 22 could potentially apply. However, any claim would first depend on establishing that the output itself is an infringing copy, which remains a significant hurdle as outlined above.

Possessing or dealing with infringing copies

Section 23 of the Copyright Designs and Patents Act 1988 could apply where a person who possesses or distributes an AI-generated output reproduces a substantial part of the Gruffalo illustrations in the course of business, knowing or having reason to believe it is infringing, may constitute secondary infringement. As with section 22, the key hurdle would be establishing that the AI-generated output is itself an infringing copy.

Providing means for making infringing copies

Section 24 of the Copyright Designs and Patents Act 1988 may be the most useful where an AI developer supplies a system capable of generating outputs that reproduce a substantial part of the Gruffalo illustrations. Scheffler could argue that the provider has supplied the means of making infringing copies, particularly if it knew or had reason to believe the system was being used for that purpose.

As above, liability is likely to depend on whether the outputs are themselves infringing copies and whether the AI developer has requisite level of knowledge that their platform would or could be used to produce infringing copies. Nevertheless, without reform of the Copyright Designs and Patents Act 1988 to account for the rapid development of AI capabilities, this is likely the best route to limiting copyright infringement through AI platforms as it targets the source as opposed to the end user.

Future hurdles for AI copyright claims in the UK

Ultimately, while the threatened claim raises a number of novel and interesting questions under UK copyright law, any litigation would face significant evidential and practical hurdles. Issues relating to subsistence, infringement and, importantly, identifying an appropriate defendant are likely to make such claims complex and uncertain. As a result, if the Gruffalo does go to court, it may prove as significant for the legal questions it raises about AI training and output generation in the future.

For more information, please contact our intellectual property team.

Sign up for legal insights

We produce a range of insights and publications to help keep our clients up-to-date with legal and sector developments.  

Sign up