Alex Bisset considers the significance of the recent ruling in Getty Images and Others v Stability AI Ltd [2025] EWHC 38 (Ch) in complex Artificial Intelligence disputes
Introduction
In 2016, a group of professionals from various organisations, including Microsoft, developed a project they titled “The Next Rembrandt”. Using an AI machine learning platform which analysed all 346 of Rembrandt’s paintings, the project developed entirely new, AI generated Rembrandt-style painting over 350 years after the artist’s passing. The project raised interesting legal and ethical questions: should the resultant painting be protected by copyright?
Several years on from ‘The Next Rembrandt’, a highly novel project at the time, generative AI has seen a meteoric rise. In conjunction with this rise, the Courts of England and Wales have already begun to see high profile litigation concerning similar legal and ethical questions that ‘The Next Rembrandt’ posed. In the space of under a decade, what were once trivial questions and interesting thought exercises for lawyers have now become flashpoints of contentious, very real litigation, to which answers are highly anticipated.
Enter Getty Images, one of the largest names in digital media, supplying a library of over 477 million assets; and Stability AI, developer of the open source generative AI model ‘Stable Diffusion’, one of the most notable AI models of the generative AI era.
Dispute Background
Getty Images (US) Incorporated, and four additional claimants within the Getty Images group of companies, brought an intellectual property-based claim against Stability AI, primarily (but not solely) based on copyright and database right infringement. Specifically, it was alleged that the Defendant had “scraped” millions of images from several of Getty Images’ websites without the Claimants’ consent, and used those images unlawfully as input to train and develop Stable Diffusion. In addition, it was further claimed that generated images output by Stable Diffusion were infringing, as they reproduced a substantial part of the Claimants’ copyrighted works.
The Sixth Claimant in the proceedings is Thomas M Barwick Incorporated, of which Mr Thomas Barwick, a photographer, is director. Content produced by the Sixth Claimant was licensed to Getty Images, and many thousands of his images and photos are available on Getty Images’ websites.
The Claimants sought, pursuant to CPR 19.8(1), an order that the Sixth Claimant may act as representative for a class of persons who were “owners of the copyright subsisting in artistic works and film works that have been licensed on an exclusive basis to the First Claimant, the copyright in which has been infringed by the Defendant” [16].
The representative claim was sought on the basis that it would be disproportionate to identify all members of the class owing to their enormous number (estimated to be around 50,000). This point was disputed by the Defendant. However, the Defendant did admit that least “some” images from Getty Images’ websites had been used to train Stable Diffusion [8].
Each copyright owner was alleged to have a concurrent right of action with Getty Images (US) Incorporated under sections 101 and 102 of the Copyright, Designs and Patents Act 1988.
Mrs Justice Joanna Smith DBE, the Judge, noted that the issues in the case were “highly complex” and that it was “a challenge on both sides to work out (the issues) and to plead how they are to be addressed” [47].
Judgment
The two key issues for the Court to determine were: 1) whether the Sixth Claimant had the same interest as the represented parties, and 2) whether the class to which they belonged could be adequately defined.
In considering these points, the Judge dismissed the Claimants’ representative claim, although the Sixth Claimant was permitted to remain as a claimant in the proceedings.
As to the first key issue, the Judge preliminarily considered that the question of whether members of the class have “the same interest” can only be addressed on the assumption that it is clear that any particular person qualifies for membership of the class [67]. The Judge ultimately came to the view that the Claimants had failed to identify a “core proposition” (per the recent judgment of Nugee LJ in Commission Recovery Limited v Marks & Clerk LLP [2024] EWCA Civ 9), or that the proceedings would give rise to a declaration which would be equally beneficial to every member of the class; where there is no benefit to all members of the class, there cannot be the same interest in the claim [92].
As to the second key issue, the Judge considered that the proposed class definition was indeed entirely dependent on the outcome of proceedings, and thereby fell afoul of the principle in Emerald Supplies Ltd v British Airways Plc [2011] Ch 345: namely, that the membership of the class should not depend on the outcome of litigation [73].
The Judge took further issue with the class definition, in that there was no definitive list of the copyright works alleged to have been used to train Stable Diffusion [77]. On this point, the following was noted by the Judge:
“There seems to be no way at present to identify the members of the class… the enormous number of images in the Training Dataset(s), as well as the enormous number of Copyright Works in issue, mean that in practice the exercise of trying to determine which works within the Training Dataset(s) are Copyright Works ‘would be wholly disproportionate and practically impossible without significant resources.’” [79]
The Claimants had, in the alternative, sought permission pursuant to CPR 19.3(1) to proceed with their Claim without joinder of the copyright holders licensing their works to Getty Images whose works were alleged to have been infringed. The Judge, however, refused permission for this on the basis that the Claimant had provided insufficient evidence as to the prejudice the Defendant may be exposed to by the enormous number of potential future claims by copyright holders.
Procedurally, the Judge considered that “a pragmatic way forward must be found which does not involve the joinder of 50,000 potential claimants” [119], and that it may be possible for the issues arising out of the Claim to be dealt with by a sample of works [120].
Analysis
Although only a procedural judgment, there is much that the High Court’s decision can tell us about litigation involving generative AI. It is plain that the enormous number of copyrighted works alleged to have been infringed, and the potential number of Claimants, make case management of infringement claims involving the ‘scraping’ of images by AI models a highly challenging task.
The significant complexity of the case management required in this case in many ways underscores the complexity of the underlying facts and issues. Indeed, it is not difficult to see why Claimants aver that it would be disproportionate to identify and join each and every one of the approx. 50,000 individual copyright holders affected by the Defendant’s activities, and why, in the circumstances, a representative Claim was sought.
The Court’s refusal to proceed with the representative claim is, however, also understandable; it would be arguably tantamount to declaring that even where copyright holders and the works infringed cannot be individually identified, there would nonetheless be a chance of succeeding in claim for copyright infringement; a finding which would be hugely advantageous to the Claimants.
In light of the huge public interest in the ultimate outcome of such proceedings, any interim decision which may be seen as unduly favouring one side over the other is likely to have substantial ripple effects even before the final decision is announced. Both in the interim and at the conclusion of the proceedings, it is apparent that the Court will wish to strike a careful balance between artistic and technological freedom, and the rights of copyright holders.
The case of Getty Images v Stability AI provides an interesting insight into the case management of a highly complex, emerging area of dispute. It is clear that high-level litigation concerning generative AI models is in its ‘wild west’ phase, and that Courts are still grappling with ways to make litigation in this area feasible. The future of the dispute between Getty Images and Stability AI remains to be determined for now. It may well be possible that a ‘sample’ approach as suggested by the Judge is adopted, which would allow the key issues to be decided without an extensive analysis of each alleged infringement.
Closing Points
The ramifications of any final decision are likely to be enormous. On one hand, a ruling overly in favour of Getty Images may lead to accusations of stifling the creative and technological freedom of those developing generative AI models. On the other hand, a ruling overly in favour of Stability AI may lead to accusations of an ‘open season’ against the works of copyright holders distributed by web-based databases, so long as one uses an AI model to scrape the images. Further to this, there is the potential for the very foundations of copyright law to be questioned and revised by lawmakers in response to the rapid change in circumstances created by generative AI.
It is not surprising in such circumstances that the High Court has adopted a careful procedural approach leading up to the final determination of the issues. But it is not difficult to imagine that even after a final decision is granted by the High Court, there is distinct potential for further appeals to be made on some, if not all, the issues in dispute.
Should copyright subsist in the 347th painting of Rembrandt? That question remains unanswered for now. But an answer feels much closer now than it did in 2016.