Class Dismissed? Representative Claims in Getty v. Stability AI

Earlier this year, the English High Court considered an application for strike out of a representative action in the ongoing dispute between Getty and Stability AI. The case is at the intersection of intellectual property (IP) issues in connection with generative artificial intelligence (AI) and class actions in the UK – issues close to Cooley’s heart – and we have been following developments with interest. The trial is listed to be heard in June, and so we’ll likely need to wait a while longer before we receive a decision on the IP implications arising from the use of third-party materials to train generative AI models. However, this recent decision[1] provides insight into the extent to which the English court will be prepared to entertain class actions for IP claims, even in respect of complex tech litigation.


Background

In summary, in January 2023, Getty (the licensee of millions of photos and videos) issued a claim against Stability AI (the provider of an AI model for text-to-image generation) alleging copyright infringement, database right infringement, trademark infringement and passing off. In essence, Getty alleged that Stability AI had been using its content on a massive scale, without permission, to train and develop its AI model, and that such use constitutes an infringement of its IP rights.

The representative claim

Getty had, from the start of the proceedings, sought to rely on a ‘representative claim’. This meant that one of the claimants – a company called Thomas M. Barwick Inc. (which exclusively licensed content to Getty) – was acting as a representative for other potential claimants that had an interest in the proceedings, i.e., other parties that had granted Getty an exclusive licence and whose copyright had allegedly been infringed.

A representative claim (under CPR 19.8) is one mechanism under English law that enables multiple claimants to bring a claim under one set of proceedings. Multiple claimants that have ‘the same interest in a claim’ as the representative can participate in the action, and the judgment of the court is binding on all persons represented in the claim. This reflects the fact that it is not always practically convenient to join all interested persons to a claim, especially if there are many (such as in this case, where there were around 50,000).

Whilst the representative claim initially went under the radar somewhat during case management, once the trial date was fixed, Stability AI applied for the representative claim to be struck out on the basis that there was an unclear and imprecise definition of the represented class. Stability AI argued that insufficient thought had been given to how members of the represented class could be identified without an individualised assessment. Further, as the proposed class was defined by reference to those whose copyright had been infringed, it was argued that the class could not be identified prior to the outcome of the trial on liability – a fatal flaw for a representative claim.

The High Court’s judgment

The court sided with Stability AI and concluded that, as things stood, there was no way to identify the common interest amongst the members of the class represented by Thomas M. Barwick Inc. Due to the technical complexity and vast scope of the alleged infringement, the identification of the members who had exclusively licensed content to Getty and whose content had been used to train and develop Stability AI’s model was practically impossible without significant resources. In the absence of clear proposals from the parties as to how these obstacles could be overcome, the representative claim could not continue – especially considering the swiftly advancing trial date.

However, the court did not completely shut the door on the claim continuing as a class action. Although the representative action (under CPR 19.8) could not continue, the court suggested that the problems identified were not intractable, and that it would be possible, should an appropriate application be made by Getty (with evidence), for interested parties to be included without joinder under CPR 19.3 and section 102(1) of the Copyright, Designs and Patent Act 1998 (CDPA). Such an approach would avoid the need for Getty to demonstrate that the parties have ‘the same interest in the claim’. This would open the door for issues of infringement to be dealt with by sample, avoid joining 50,000 potential claimants to the proceeding, and allow the court to focus on the ‘big picture’ issues. However, Getty would need to provide Stability AI with comfort that it would be protected from future claims for relief by exclusive licensors that had not been made party to the proceedings. This, the court suggested, could be easily satisfied if Getty were prepared to provide an undertaking to indemnify Stability AI against that risk.

Getty did not provide such an undertaking at the application hearing, but did so shortly afterwards. The court subsequently ordered that Getty’s claim on behalf of those other parties that had granted Getty an exclusive licence could continue, without them being joined as parties pursuant to CPR 19.3 and section 102(1) of the CDPA.

Takeaways

This case illuminates the difficulties in pursuing a representative claim in circumstances where the identification of class members is complicated by complex technical issues – such as IP infringement in the context of AI models.  

If a representative claim is pursued, the scope of that claim, and the ramifications for trial, should be considered in detail from an early stage of the proceedings. In this case, the court suggested that had Getty given more thought to the representative claim under CPR 19.8 earlier on, it may well have survived.

Crucially, the decision highlights the court’s willingness to be flexible and pragmatic in allowing class actions to proceed, even in technically complex cases. The English courts are willing to take an active role in the case management of mass claims, provided that the certainty and fairness for the defendant is not prejudiced as a result.    


[1]  [2025] EWHC 38 (Ch).

Contributors

Alex Radcliffe

Andrew Linch