Blurring the “big picture”: latest decision in Getty v Stability AI highlights challenges in complex representative claims

As we enter 2025, buzz around AI (and generative AI in particular) shows no signs of slowing down. As a result, Getty v Stability AI is being followed with great interest in the UK as a key test case for how IP infringement claims against use of unlicensed training data by AI model developers might be handled. 

As we have reported on previously (see here and here), Getty alleges that Stability AI’s generative AI image generator “Stable Diffusion” was trained using images scraped from its websites without permission, thereby infringing its copyright and database rights. It also alleges that the outputs produced by Stable Diffusion reproduce substantial parts of Getty’s works and/or bear its protected water marks, and therefore also infringe. 

While the case clearly has significant IP ramifications, the latest instalment in the saga (a recent High Court decision on a procedural application) shines a light on the procedural and practical challenges that a novel copyright claim of this size can give rise to. 

Representative claim

The decision primarily deals with whether the Sixth Claimant (Thomas M Barwick Inc, which has exclusively licensed images to Getty) could act, per CPR 19.8, as a representative for a class of around 50,000 individual copyright owners who had exclusively licensed works to Getty and whose copyright (so Getty allege) has been infringed by Stability AI. 

As the judge noted, “the development of digital technologies has added to the potential for mass harm for which legal redress may be sought” and representative actions have come to be seen as a flexible tool by the courts to allow for the administration of justice in those circumstances (even if the test for when they apply imposes a fairly high bar). 

In coming to its decision, the court had to consider, firstly, whether the case was within the rule at all – i.e. whether the class could be satisfactorily identified in practice – and, if so, whether the Sixth Claimant had the same interest (and no conflict of interest) with the represented parties in that class. Ultimately, the court found that the class definition put forward by Getty was too uncertain as identification relied on the result of the litigation (i.e. whether a person’s copyright had been infringed). Even if the class definition was amended to remove that element such that it covered all exclusive licensors, under specifically identified agreements, of works which were used to train Stable Diffusion, the court found it would still not be possible (without significant resource) to determine which copyright works had been used to train Stable Diffusion in the UK (and therefore to identify the relevant licensors) and the parties had put forward no concrete proposal on how that determination might be achieved. 

Even if it was wrong on these points, the court was not persuaded that it should exercise its discretion to permit the representative claim as doing so would not improve or simplify existing uncertainties surrounding the management of the case. Therefore, the court accepted Stability AI’s application that the Sixth Claimant could not act as representative. 

The court also considered an alternative proposal to proceed on the basis that the class of exclusive licensors were not joined to the action further to s.102 of the Copyright, Designs and Patents Act 1988 and CPR 19.3. Those provisions give the court discretion to order that persons with concurrent rights of action need not be joined as a party to the proceedings. In determining whether to exercise that discretion, the key question for the court was whether Stability AI would suffer prejudice if such an approach was ordered. Ultimately, this proposal was also dismissed on the basis that sufficient evidence had not been adduced, nor comfort given, that Stability AI would be protected from future claims for the same relief by exclusive licensors that had not been made party to the proceedings. Nevertheless, the judgment left the door open to such an application if supported by appropriate evidence, noting that it would have the benefit of allowing the focus of the trial to be on the “big picture” issues (which could potentially be resolved by analysis of a sample of works) rather than wasting court time on debates over things like the interpretation of individual exclusive licences and whether they effectively prevent the relevant licensors from bringing proceedings. 

Looking ahead

On the specific issue of representative action, the decision is a useful reminder for those seeking to rely on CPR 19.8 of the importance of appropriately formulating the class definition and ensuring that the representative action ties into the broader approach to case management. The courts will be loathe to use their discretion if the use of representatives complicates rather than simplifies proceedings. However, in leaving the door open to a further application, it does seem that the court is keen to find a solution that will allow the core issues in the case to be heard without joining 50,000 potential claimants and, to achieve this, it appears to favour the approach of not joining the exclusive licensors to the action under CPR 19.3. That though would likely require Getty to provide evidence directed to, or comfort (e.g. an indemnity) to Stability AI in respect of, future claims by those licensors.

More broadly, the judgment gives some insight into the various disclosure and case management challenges the parties are facing, given the enormous number of images in the training dataset and potential authors of affected copyrightable works. It seems that some fairly fundamental questions have yet to be answered, in particular what set of realistic and manageable issues will be taken forward to trial, the evidential basis on which those will be assessed (e.g. the approach to sampling and extrapolation) and even the identity of the claimants. This, in part, is because the parties are struggling to agree on procedural matters, likely for fear of disadvantaging themselves. It is therefore clear that the court is going to have to take a very active role in case management to encourage the parties to adopt a pragmatic approach to these challenges. Further CMCs are expected early this year and, with trial set for 9 June 2025, there will no doubt be further important practical and procedural issues that need resolving in the coming months before we get an answer to the “big picture points”.