On November 4 2025, the UK High Court handed down its much-anticipated decision in Getty Images v Stability AI. Whilst the scope of the case ended up being much narrower than initially expected, after Getty dropped its claims for primary copyright and database right infringement at the end of the trial, the court’s views on Getty’s claims of secondary copyright infringement were keenly awaited.
As most readers of Managing IP will be aware by now, the court rejected those copyright claims. Getty did secure a sliver of success on its trademark infringement claims, but the practical reach of that win was modest despite the word count that those issues occupied in the judgment.
As the dust starts to settle, this article explores the core reasoning behind the copyright aspects of the court’s decision, what it might mean in practice for AI developers and UK rights holders, and which aspects of the decision Getty might look to appeal.
Background
Stability AI has developed a generative AI model known as Stable Diffusion that creates synthetic images in response to prompts entered by users. Getty alleged that Stable Diffusion was trained on millions of copyright-protected images that were scraped from its website without its consent. Whilst Getty had originally asserted that those actions infringed its UK copyright and database rights and that the resulting images produced by Stable Diffusion reproduced substantial parts of its copyright-protected works, Getty dropped those claims of primary infringement during the trial, largely as a result of evidential and territoriality issues.
From a copyright perspective, that left Getty with a claim for secondary infringement only.
Secondary infringement of UK copyright
Secondary infringement is targeted at downstream dealings with copyright-infringing goods. Sections 22 and 23 of the Copyright, Designs and Patents Act 1988 (CDPA) provide that importing or dealing with an article that is, and which the person importing or dealing with that article knows or has reason to believe is, an infringing copy of a work is an act of secondary infringement.
“Infringing copy” is defined in Section 27 of the CDPA. Importantly, for these purposes, Section 27(3) states that “[a]n article is also an infringing copy if […] (a) it has been […] imported into the United Kingdom, and (b) its making in the United Kingdom would have constituted an infringement of the copyright in the work in question…”.
Relying on these provisions, Getty argued that:
Stable Diffusion is an “article”;
Stable Diffusion is an “infringing copy” because it has been imported (downloaded) into the UK and its making in the UK would have constituted an infringement of Getty’s copyright (although it did not argue that the model itself is, or includes, a copy of any of its copyright works, relying instead on an interpretation that Section 27(3) does not require a copy); and
Importing Stable Diffusion (through downloading in the UK) and distributing it through the Hugging Face platform amounted to secondary acts of copyright infringement.
This led to two core questions for the court to determine:
Firstly, whether the word “article” in sections 22 and 23 is broad enough to capture intangibles. Applying the ‘always speaking’ principle of interpretation (i.e., that legislation should be read as taking into account changes that have occurred since the legislation was put in place), the court agreed with Getty that it is.
Secondly, whether Stable Diffusion is capable of being an infringing copy. Here, the court ultimately rejected Getty’s argument around the letter of the legislation on the ground that, by its nature, an infringing copy must be exactly that – a copy. Stable Diffusion was found to have never contained or stored copies of any of Getty’s copyright works. It was not, therefore, an infringing copy and Getty’s claim for secondary copyright infringement failed on that basis. The fact that copies of Getty’s works may have been used in developing that model did not change that.
Practical impact of the decision
Whilst this is only a first-instance decision and remains open to appeal (see below), the immediate impact of the court’s finding on secondary copyright infringement is that importing a pre-trained generative AI model such as Stable Diffusion into the UK will not amount to secondary infringement of UK copyright where the model was trained abroad and does not store copies of the relevant copyright work within the model weights.
If that is right, then it is a good result for AI developers (the majority of whom are based outside the UK). However, it will be seen as a severe loss for owners of UK copyrights, as well as, perhaps, for the UK government.
For rights holders, this decision may be seen as holding the door open for AI developers to train their models in AI-friendly jurisdictions and then import the fruits of their labour into the UK, without running any secondary infringement risk.
As far as the UK government is concerned, the perception might be that this decision encourages AI developers to continue their training and development activities outside the UK, only accessing the UK market at a later stage, once the model is ready. Incentivising that import-only model appears to run counter to the government’s ambitions to be a global leader in AI, as set out in its AI Opportunities Action Plan (in which the UK prime minister, Sir Keir Starmer, expressly stated in his foreword that he wants the UK to be making AI breakthroughs rather than importing them from other countries).
The decision was not all doom and gloom for copyright holders, however. The court’s findings that an “article” for secondary infringement purposes can cover intangibles, together with its confirmation that downloads can amount to importation, will be welcomed by copyright holders and will have broader application outside the AI space.
The court’s views on who is responsible for (trademark) infringing outputs is also worthy of note, with the court finding that in this case it was Stability AI, not the users of Stable Diffusion, that was legally responsible.
Whilst the images in question may have resulted from a user’s prompts, ultimately the outputs were based on the functionality of the model. That functionality, in turn, depends on the training data, which was chosen by Stability AI. Admittedly, that was determined in the context of the trademark infringement aspects of the case and its future application will likely be fact-dependent (with the court leaving scope for users to be held responsible if they were to deliberately seek to generate infringing outputs); however, it can be seen how rights holders might seek to argue that the rationale is broad enough to apply to copyright-infringing outputs, too.
Indeed, only a week after the Getty decision was handed down, the Munich Regional Court found, in GEMA v OpenAI, that OpenAI was legally responsible for outputs generated using simple prompts that were found to infringe German copyright based on similar reasoning.
Potential grounds for appeal
The key question on everyone’s lips now is whether Getty will appeal and if so, on what grounds. Given Getty’s standing as a significant UK rights holder and the fact it decided to bring this claim in the UK in the first place, it would not be surprising if it does bring an appeal.
If it were to do so, the obvious part of the judgment to challenge is the court’s finding on the meaning of “infringing copy”. This, in turn, could be broken down into two core parts. Firstly, did the High Court give sufficient weight to the explicit wording of Section 27(3) of the CDPA in reaching its conclusion that an “infringing copy” has to be a copy? Secondly, even if it did, was the court right to conclude that Stable Diffusion was not a copy of, nor did it contain copies of, Getty’s works?
On this second question, copyright holders in general will seek comfort from the recent decision in GEMA v OpenAI, which found (albeit in a primary infringement context) that another generative AI model (ChatGPT) did “memorise” and “reproducibly contain” certain song lyrics, thereby infringing copyright.
However, it looks like this line of argument might not be open to Getty itself. Whilst the experts in the Getty case did acknowledge that Stable Diffusion could produce “images that are nearly identical (a memorized image)”, Getty did not actually assert that Stable Diffusion contains copies of any of Getty’s copyright works, nor did it provide any evidence of its copyright works having been memorised by the model (relying instead on the assertion that no such copy is required for secondary infringement, given the language of Section 27(3)). We can only wonder whether the court would have reached a different conclusion if Getty had sought to plead these points.
What was not covered by the decision
Whilst this first-instance decision largely went against Getty, the scope of the claims ultimately made were much narrower than initially filed.
In the end, the court did not have to grapple with any questions of whether training and developing a generative AI model in the UK using unlicensed third-party content can amount to primary infringement of copyright or database rights, nor did the court need to consider whether any exceptions to infringement might apply.
Questions relating to whether generative AI outputs can infringe copyright (either through copying or communication to the public) also did not arise. Nor were any arguments made around breach of contract (such as where the terms and conditions for websites from which training data is taken prohibit web-crawling or scraping).
Even for those questions that the court did consider, as noted above, Getty did not assert that – and the court did not therefore have to grapple with counterarguments about whether – Stable Diffusion was, in fact, a copy of, or contained copies of, any of Getty’s copyright works.
Given the above, it is fair to say that the Getty decision only really ‘scraped’ the surface on these sorts of claims. Whilst it is useful to have this decision and to hear the court’s views on secondary copyright infringement (as well as trademark infringement), there is plenty still to come.
So, what’s next?
Getty has until 21 days after the consequentials hearing (expected in the week commencing December 15) to decide whether it wishes to appeal.
In the meantime, Getty’s parallel claims against Stability AI for primary copyright infringement are continuing in the US. It will be interesting to see how that plays out and whether Stability AI’s use of Getty’s images in training Stable Diffusion in the US will be deemed to be fair use.
More broadly, in the UK, we are still awaiting the government’s response to its consultation on copyright and AI, which should include details of how it intends to balance the rights of AI developers and rights holders when training generative AI (including whether to introduce any broader exceptions to copyright and database right infringement for AI developers).
We hope to also hear the government’s views on a number of other topics covered by the consultation, such as how AI models that have been trained in other jurisdictions should be treated (which is particularly relevant in light of the Getty decision) and infringement and liability relating to AI-generated outputs.
Closely linked to that, the UK government is also due to publish, by March 2026, two reports under the Data (Use and Access) Act 2025:
An economic impact assessment of the four policy options it put forward in its consultation on copyright and AI relating to how best to balance the rights of AI developers and copyright holders when training generative AI; and
A more general report on the use of copyright works in the development of AI systems.
A progress report is due in December 2025.
Further afield, a number of court decisions are awaited in copyright and generative AI disputes in the EU, including two decisions in Germany (GEMA v Suno AI, and the outcome of the appeal in the Robert Kneschke v LAION case), as well as, potentially, an appeal in the GEMA v OpenAI case referred to above. The outcome of SNE’s claim against Meta in France is also awaited, as is the Court of Justice of the European Union’s decision in Google v Like Company (expected towards the end of 2026 or early 2027).
With the outcome of many of these cases expected to go to appeal given what is at stake, it feels like this is only the start of the journey, with plenty for intellectual property lawyers to look forward to.
