In November of last year, one of the most hotly anticipated judgments was handed down by the Courts in the case of Getty Images v Stability AI. This was the first major ruling which considered how trademark and copyright laws apply to the use of generative AI models in the UK. The judgment was delivered by Mrs Justice Joanna Smith DBE, and the lengthy 205-page document was indicative of numerous issues raised in the claim, many of which were being considered by the Court for the first time. As discussed further below, the Court rejected the majority of Getty’s claim however the tech industry and creatives alike will be watching closely as permission to appeal the decision has been granted.
Background
Getty Images (“Getty”) is a well-known provider of licensed photographic content, and discovered that some of its images seemed to appear in datasets that had been used to certain versions of Stable Diffusion, an imagine generation model developed by Stability AI. Getty brought a claim against Stability AI over the use of their database of images and associated text in the training of the AI model, as well as the outputs of that model, which seemed to resemble their content and even included versions of the famous Getty watermark.
As the claim progressed, however, Getty’s claim considerably narrowed due to the lack of evidence that Stable Diffusion was trained in the UK. Getty therefore abandoned its copyright infringement claims, which require a territorial element in the UK. This means that the Court is still yet to decide on the key issue of whether training and developing AI models in the UK on images protected by copyright constitutes infringement.
Two main claims were therefore left to be decided:
- Secondary copyright infringement, under sections 22, 23 and 27(3) Copyright Designs and Patents Act 1988 (“CDPA”); and
- Trade mark infringement, under sections 10(1), 10(2) and 10(3) of the Trade Marks Act 1994.
Secondary Copyright Infringement
Getty contented that, pursuant to section 27(3) CDPA, Stable Diffusion is an “infringing copy” under the CDPA because it has been imported into the UK (through being downloaded) and its distribution in the course of business. The term “infringing copy” is defined in s27 CDPA and a key issue in the judgment is whether Stable Diffusion was “an article” for the purposes of this section. The term “article” is not defined in the statute, and after considering the issue the judge agreed with Getty that the term “article” includes intangible information, following the principle that statute should be interpreted taking into account changes have occurred since the statute was enacted.
Despite this, the Court held that the AI model weights are not a “copy” for the purposes of the CDPA, as it must be a reproduction of a copyrighted work. The Judge held that, despite the model weights being altered by training using works protected by copyright, by the end of the training process the model itself does not store any of those copyright works and are not themselves an infringing copy, but “purely the product of the patterns and features which they have learnt over time during the training process”.
Trade mark infringement
Getty alleged that Stable Diffusion generated images that included their watermarks, or distorted versions of them, in some of its outputs in early versions of the AI model.
The Court found partially in favour of Getty in respect of the trade mark infringement claim, finding that the watermarks generated were trade mark infringement pursuant to 10(2) TMA although the judge highlighted that her findings were “both historic and extremely limited in scope”. She noted that it was impossible to know how many, and on what scale that the watermarks would have been generated in real life by users.
The judgment was likely to be disappointing for artists and creatives, whose copyrighted works are often used in the training and development of LLMs (Large Language Models). Developers of AI models will likely welcome the court’s decision, particularly in rejecting the argument that model parameters themselves are infringing.
Appeal
There is still a great deal of uncertainty for rights holders and developers alike as to what the future of AI litigation may look like. Developers of AI may take heart from the decision however may want to mindful of where any models are trained and consider training models outside of the UK until there is more certainty as to the position. On 16 December 2025, the Judge granted permission for Getty to appeal her judgment in relation to the secondary infringement claim, and so many will be closely following to see if the Court of Appeal will affirm the High Court’s decision.