Getty v Stability AI - first AI judgment made but key questions remain unanswered

The judgment was passed down on 4 November 2025 in Getty v Stability AI, the UK’s first case on infringement of intellectual property by generative AI.

Well over half of the 205-page written judgment deals with trade mark infringement claims, upholding certain - but by no means all - aspects of Getty’s claims in this regard. Given that evidence showed that certain Stability models would reproduce proprietary watermarks on synthetic images, this is perhaps not surprising and confirms that AI developers can be liable for outputs that misuse protected marks.

What remained of the copyright aspects – secondary infringement – comprises just a small section of the judgment, although this is where the decision does get more interesting. In particular, the Judge held that the term ‘article’ in the relevant copyright legislation can indeed include something that is intangible, such as an electronic copy stored in an intangible medium. This is the first time since the relevant law’s enactment in 1988 that this has been considered.

However, having held that an intangible article can be an ‘article’ under the legislation and so can in principle comprise an ‘infringing copy’, the Judge proceeded to dismiss Getty Images’ claim of secondary copyright infringement for importing, possessing and dealing with an alleged ‘infringing copy’.  She held that the relevant legislative provisions (section 27 of the Copyright, Designs and Patents Act 1988), when read in the context of the whole Act, require the ‘article’ to itself be a copy of the original work in order to be an ‘infringing copy’.  Because an AI model such as Stable Diffusion does not actually store or reproduce any actual copyright works – and has never done so (which was agreed by the parties’ experts) – it cannot be an “infringing copy” under the relevant legislation, meaning Getty’s claims for secondary copyright infringement must fail.

It would not be sufficient for the making of the AI model to involve copyright infringement (this issue was outside the court’s jurisdiction) because the AI model itself – although an ‘article’ – was not a copy. The case on primary infringement was dropped at trial, there being insufficient evidence that any training of the AI model took place in the UK.

Although Stability AI largely prevailed, this was not a sweeping win for either side, nor a decision that will have the impact some may have hoped for. Fundamental issues concerning whether the training of a gen AI model using third party copyright works amounts to copyright infringement, whether there was primary copyright infringement by reason of the gen AI outputs, and the database rights claims were not proceeded with at trial.

In the absence of a decision on primary infringement, it is not clear what remedies would be available to copyright holders if the training of gen AI model did take place in the UK. Important questions remain. Even if the training and development of the model amounted to copyright infringement, would the courts order an injunction against the use of an AI model if the model itself does not and seemingly cannot infringe copyright in the original work? Would there be any requirement to retrain such model in light of this judgment?

Also notable is that Getty was not itself an owner of the copyright works but a licensee under licences governed by New York law. While an exclusive licensee would have concurrent rights with the copyright owners and be jointly entitled to remedies, depending on the wording, some of the licences were not considered exclusive as they were granted to multiple entities (which is permitted under New York law but made the licence non-exclusive under English law).

Getty’s US case against Stability is still ongoing and in the UK, this may not be the end of the story if Getty decides to appeal.