With its decision of 4 November 2025 in Getty Images (US) Inc. & Others v. Stability AI Ltd., the High Court of Justice in London examined legal issues arising from the use of copyright-protected works and trademarks in the training and operation of generative artificial intelligence systems. The Court analysed the dispute between the owner of an extensive catalogue of images (Getty Images) and the developer of the “Stable Diffusion” generative model (Stability AI), ruling on two key questions: whether secondary copyright infringement could be established and whether trademark protection applies to generative outputs. Presided over by Mrs Justice Joanna Smith, the Court dismissed the allegation of secondary copyright infringement, rejecting the argument that the trained model constitutes an “infringing copy” of the works used during training. It instead partially upheld the trademark claim, finding that the synthetic generation of watermarks can amount to an infringement. The judgment traces an initial legal boundary while deliberately leaving unresolved substantive issues such as the lawfulness of the training process, deferring the definition of a broader systemic balance to future developments. The dispute arose from an action brought by the Getty Images group companies against Stability AI. Getty Images, owner of a vast catalogue of protected images, claimed that Stability AI had unlawfully used millions of such works to train its “Stable Diffusion” model. The original claims included primary and secondary copyright infringement, database rights infringement and trademark infringement. The proceedings were significantly narrowed when Getty Images withdrew the allegations of primary copyright infringement and database rights infringement, due to the absence of sufficient evidence showing that the scraping and training activities had taken place in the United Kingdom. Consequently, the case focused exclusively on secondary copyright infringement and trademark infringement, limiting the analysis to the distribution and use of the already trained model within the UK. Getty Images’ core argument on copyright relied on an extensive interpretation of the Copyright, Designs and Patents Act 1988 (CDPA). It claimed that the Stable Diffusion model, once trained, constituted an “article” that is an “infringing copy”, because the creation of its parameters through training on protected works would have amounted to an act of unlawful copying had it occurred in the UK. The distribution of such an “article” within the UK would therefore amount to secondary infringement. The Court recognised that the notion of “article” under the CDPA can include intangible goods, such as an AI model. However, it rejected Getty Images’ argument, clarifying that to qualify as an “infringing copy”, the article must contain or store a reproduction of the original work. The Court found that the Stable Diffusion model does not store the images used for training; its parameters reflect the output of patterns and abstract features learned. The model encodes a statistical representation of concepts, without incorporating the specific works. Consequently, the Court concluded that Stable Diffusion is not an “infringing copy” and fully dismissed the claim of secondary copyright infringement. Getty Images did obtain legal recognition on the separate issue of trademark infringement, governed by the Trade Marks Act 1994 (TMA). The Court found that certain earlier versions of Stable Diffusion could generate synthetic images containing the “Getty Images” and “iStock” watermarks, identical to the registered trademarks. Mrs Justice Smith held that this synthetic generation of identical signs was potentially capable of giving rise to consumer confusion, amounting to an infringement under section 10(2) TMA. However, the Court strictly limited the scope of this finding, describing the infringements as “historic and extremely narrow in scope” and noting that the issue had been remedied in later model versions. Furthermore, Getty Images failed to prove unfair advantage or harm to the reputation of the marks, which are required for aggravated infringement under section 10(3) TMA. The judgment therefore confirms the principle that trademark protection applies to the generation of confusingly similar signs by AI, but denies any significant compensatory consequences in the case at hand. The Getty Images v. Stability AI judgment is a useful, albeit non-definitive, indicator of the direction that the legal debate on generative AI may take. It provides a precedent in UK law that, at this stage, favours the distribution of technological innovation by establishing that the final model is not, in itself, an unlawful copy. However, this approach is neither universal nor conclusive. The decision intentionally avoids the most contentious issues — the lawfulness of training and liability for substantially similar outputs — and may diverge from approaches emerging in other key jurisdictions, such as the European Union or the United States. Rather than drawing definitive boundaries, the decision sketches the perimeter of a new legal battleground, which will be shaped by future litigation and regulatory interventions seeking a still-unsettled balance between intellectual property protection and AI development.
2026 - Morri Rossetti
