The High Court has ruled against an application to bring a representative action on behalf of over 50,000 copyright holders in the much-publicised case involving Getty Images and Stability AI.

Background

The application was part of a much larger and highly consequential dispute between various companies within the Getty Images Group (“Getty”) and the UK-based AI company, Stability AI.

Stability AI has developed a deep learning generative AI model – Stable Diffusion – which generates images following prompts from its users.  Getty alleges that Stability AI has scraped millions of its images without consent and used them to train and develop its model.

Accordingly, in addition to claims for trade mark infringement, database right infringement, and passing off, Getty has brought claims of copyright infringement on three bases:

1. that during the training and development of Stable Diffusion, Getty’s copyright-protected content was downloaded on servers and/or computers in the United Kingdom;
2. that secondary infringement of copyright occurred by reason of the importation of the pre-trained Stable Diffusion software into the UK; and
3. that the output of Stable Diffusion infringes copyright since it reproduces a substantial part of Getty’s copyright-protected works.

In addition to the various Getty companies issuing proceedings, a sixth defendant was added, Thomas M Barwick, Inc. (“Barwick”), in a representative capacity. As the Court describes it, Barwick is a company “engaged in the process of creating and licensing visual content” whose director, Thomas Barwick, had “a long professional relationship with the Getty Images Group”. For its part, Barwick claims that it assigned all legal and equitable title in its copyright-protected works to Getty, and that the copyright in these works was subsequently infringed by Stability AI.

Critically, Barwick also sought to act as a representative under CPR 19.8 for a class of over 50,000 copyright owners who had equally granted an exclusive licence to Getty and, so it is alleged, whose copyright was infringed by Stability AI.

In a Response to a Request For Information, the class of person for whom Barwick claimed to be a representative was set out as follows:

“…the class of persons represented by the Sixth Claimant are those who are 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. The persons that fall within such class can be identified on the basis that (i) they have entered into an exclusive licence with the First Claimant in respect of artistic works and/or film works; and (ii) the exclusively licensed works include works which were used to train Stable Diffusion, as alleged in paragraph 34 of the Particulars of Claim”.

Stability AI argued that this constituted an unclear and imprecise definition of the represented class and denied that Barwick shared the “same interest” (as required under CPR 19.8) as the other alleged owners of copyright who had exclusively licensed work to Getty.

Judgment

Mrs Joanna Smith DBE began by considering the requirements of CPR 19.8, and the most recent case law in this area, including the Supreme Court’s judgment in Lloyd v Google [2021] UKSC 50.

Drawing the various strands together, she summarised the task of the court in a case concerning CPR 19.8 as asking the following: (a) is this case within the rule at all? (b) if so, should the could nevertheless in its discretion direct that the claimant may not act as a representative? As for part (a), two further questions arise: (1) do the claims of each member of the class raise a common issue or issues? and (2) is there any relevant conflict of interest between them?

The judge turned to consider the class as defined by the Claimants, and pointed to an obvious and potentially fatal problem: in defining the class as copyright holders whose work had been infringed, the class was dependent upon the outcome of the litigation (namely whether, in fact, infringement had taken place). As the Court explained, following Lloyd v Google and Emerald Supplies Ltd v British Airways Plc [2011] Ch 345, it is a general principle that membership of the class should not depend upon the outcome of the litigation. Accordingly, the judge held that “the question of whether copyright has been infringed by the Defendant [is] a question that can only be determined at trial. Thus, if the class is dependent upon this definition, it appears to me plainly to fall foul of [this] principle”.

In order to circumvent this problem, the Claimants sought to focus on the second sentence of its definition of the class at the hearing. This merely required that a member of the class (a) entered into an exclusive licence with Getty in respect of a copyright work and (b) that copyright work was used in the training of Stable Diffusion. In the Claimants’ view, both were question of fact and not contingent on the outcome of the litigation.

However, the judge pointed out the difficulties with this argument, most notably that there is “no definitive list of the Copyright Works that have been used to train Stable Diffusion and the Defendant has only made limited admissions in its Amended Defence to the effect that “at least some” images from the Getty Images Websites and/or some Copyright Works were used in training Stable Diffusion”.

Stability AI argued that whether a given work was actually used to train Stable Diffusion would be wholly dependent on its own facts and that the Claimants would need to establish that one or more works of each author within the alleged class were in fact (a) downloaded; (b) by Stability AI; and (c) in the UK.

The judge was persuaded by these arguments, stating that “standing back, and trying to see the wood for the trees, it is plain that the question of which Copyright Works have been used for training Stable Diffusion is not a question that can currently be determined. Indeed this does not appear, at present even to be a question which the parties had anticipated would need to be resolved at the First Trial, albeit the Claimants belatedly appear to acknowledge its significance in connection with the proposed representative claim. It is also plain that the class that is intended to be defined by reference to this question cannot be identified”.

As a result, she concluded that she could “see no basis on which the court can be satisfied that any particular person qualifies as a member of the class proposed or that it therefore has jurisdiction to permit this representative claim”.

She also added that even if she was wrong on this question of whether the claim fell within the scope of CPR 19.8, she was not persuaded in the exercise of her discretion that the representative claim should proceed to trial on the information she had to date. As she explained, “the adequacy of the definition also goes to the exercise of my discretion in deciding whether it is just and convenient to allow the claim to be continued on a representative basis (see Lloyd v Google at [78]), and for the reasons I have set out above, there are serious problems around the definition of the class which play directly into the management of this case to trial”.

Having failed to persuade the judge of their position on a representative claim under CPR 19.8, at the hearing (and without a formal application) the Claimants pursued an alternative argument that “the court permit them to pursue their claims in the absence of joinder of owners of copyright-protected works with whom they have concurrent rights of action, pursuant to CPR 19.3 and s.102(1) Copyright, Designs and Patents Act 1988 (“CDPA”)”.

The judge rejected the argument from Stability AI that a failure in relation to CPR 19.8 necessarily precludes a party relying on CPR 19.3. However, she ultimately concluded that, given the that lack of a properly pleaded application with sufficient evidence, this was “simply not an application that the court can possibly accede to at this stage” and that, accordingly, granting permission udder CPR 19.3 and s.102(1) CDPA would not be consistent with the requirements of the overriding objective.

To read the judgment in full, click here.