In the Getty Images v Stability AI judgment of 4 November 2025, the High Court of England and Wales held that the model weights of Stable Diffusion are not an “infringing copy” of Getty’s photographs, because the model does not store or reproduce them. Getty’s secondary copyright claim failed and its trade mark claim succeeded only in a narrow, historic way. Getty has permission to appeal the copyright point, and if the Court of Appeal disagrees, importing or distributing AI models in the UK could become a copyright risk even when training happened abroad. This explainer is for AI developers, deployers and rights holders with UK exposure.
Key takeaways
- Getty dropped its training claim because there was no evidence that Stable Diffusion was trained in the UK.
- The court accepted that an “article” can be intangible, but held that a model which does not store or reproduce the works is not an infringing copy.
- Trade mark infringement was found only for watermarks generated by older model versions, on specific examples.
- On 17 December 2025 the trial judge gave Getty permission to appeal the secondary infringement ruling, describing the question as novel and important.
- A Munich court reached a different result on different facts in GEMA v OpenAI, where song lyrics were memorised in the model.
What did the High Court decide in Getty v Stability AI?
The judgment of Mrs Justice Joanna Smith ([2025] EWHC 2863 (Ch)) was handed down after a trial in June 2025. Shortly before closing submissions, Getty narrowed its case considerably, so the court ruled on far less than the headlines suggested.
| Claim | Outcome |
|---|---|
| Training and development in the UK (primary copyright) | Abandoned: no evidence that training took place in the UK |
| Infringing outputs (primary copyright) | Abandoned: the prompts in evidence had been blocked, so the relief was substantially achieved |
| Database right | Abandoned, given its link to the two claims above |
| Secondary infringement (importing and dealing in an infringing copy, sections 22 and 23 CDPA) | Dismissed |
| Trade marks, sections 10(1) and 10(2) of the Trade Marks Act 1994 | Partly upheld for iStock watermarks from v1.x models and a Getty Images watermark from v2.x, on specific examples; findings described as “historic and extremely limited” |
| Trade marks, section 10(3) (reputation) | Dismissed |
| Passing off | Not decided |
Are model weights “infringing copies”?
Under the Copyright, Designs and Patents Act 1988 (CDPA), importing an “article” into the UK or dealing with it in the course of business can infringe if the article is, and the defendant knows or has reason to believe it is, an “infringing copy” (sections 22 and 23). Section 27(3) treats an imported article as an infringing copy if making it in the UK would have infringed copyright.
Getty did not claim that Stable Diffusion contains copies of its works. Its argument was that the model is an infringing copy because making its weights would have infringed if done in the UK. The court accepted the first half of Getty’s legal case: an “article” can be intangible, such as a model downloaded from the internet. It rejected the second half. The judgment explains that the model weights are parameters learned during training and that the model does not store the training data; an AI model which does not store or reproduce any copyright works, and has never done so, is not an infringing copy.
The ruling turns on that factual finding about how Stable Diffusion works. A model shown to memorise and reproduce protected works would raise a different question.
What is the appeal about, and what would change?
At the consequentials hearing in December 2025, the judge granted Getty permission to appeal on all its grounds relating to secondary infringement, calling it a pure question of law that was “both novel and important”, and refused Stability permission to appeal the trade mark findings, according to CMS (21 January 2026). Getty was ordered to pay 69.4% of Stability’s costs, with an interim payment of about GBP 4.4 million.
The CMS AI case tracker lists permission as granted on 17 December 2025. We have not seen a Court of Appeal decision, so the High Court ruling stands for now. If the Court of Appeal holds that a model can be an infringing copy even when it stores no works, the consequences would be wide:
- Offering model weights for download in the UK, hosting them or bundling them into products could be secondary infringement where the training would have infringed UK copyright.
- Training outside the UK would no longer protect a model placed on the UK market.
- Knowledge would become the key battleground: once rights holders give notice, distributors may find it harder to argue they had no reason to believe the model was infringing.
How does Getty v Stability compare with the EU?
EU law takes a different route to a similar question. In GEMA v OpenAI (Munich Regional Court I, 42 O 14139/24, 11 November 2025, not final), the court found that song lyrics had been memorised in the model’s parameters and that this was a reproduction, not covered by the text and data mining exception. The facts were the reverse of Getty: memorisation was established.
| United Kingdom | European Union | |
|---|---|---|
| Key question | Is an imported model an “infringing copy”? | Is training or memorisation a reproduction, and does the mining exception apply? |
| Leading decision | Getty v Stability AI (2025): no, if the model stores no works; under appeal | GEMA v OpenAI (Germany, 2025): memorisation is reproduction; not final. A first CJEU reference, Like Company v Google (C-250/25), is pending |
| Training abroad | Primary claims need acts in the UK | AI Act providers of general-purpose models must respect EU copyright and opt-outs wherever training took place (recital 106) |
For businesses selling into both markets, the safest assumption is that the stricter result applies, and that what the model can reproduce matters as much as where it was trained.
What this means for your business
- Document where each model was trained and on what data, including licences and opt-out handling.
- Test for memorisation: can the model reproduce training works, logos or watermarks on simple prompts?
- Review UK distribution: hosting, downloads, app stores and resellers are the acts an appeal could reach.
- Check contracts with model suppliers for warranties and indemnities that cover UK secondary infringement and trade mark claims.
- If you are a rights holder, keep dated evidence of outputs and send notices that could later establish knowledge.
Our team for AI model copyright and digital asset protection can review your training, distribution and contracts across markets, and our IP enforcement and licensing team handles notices and disputes. AI Act compliance is covered within the same META Channel group.
When to bring in an adviser
- Before releasing weights in the UK or the EU. Once a model is distributed, it is hard to pull back.
- When your model outputs show logos or watermarks. Getty’s partial trade mark win rested on exactly that.
- When a supplier cannot say where or on what its model was trained.
- When you receive or plan to send a notice. Its wording can shape a later knowledge argument.
Frequently asked questions
Did the Getty v Stability judgment say AI training is lawful in the UK?
No. Getty abandoned its training claim because there was no evidence that training took place in the UK, so the court did not decide whether training on Getty’s images would infringe if carried out there. The judgment decided that importing and dealing in a model that stores no copies is not secondary infringement, plus limited trade mark points.
Is the Getty v Stability judgment final?
Not on copyright. The trial judge gave Getty permission to appeal the secondary infringement ruling in December 2025, and the appeal is pending before the Court of Appeal. Stability was refused permission to appeal the trade mark findings by the trial judge. Until the appeal is decided, the High Court ruling remains the leading UK authority.
Why did Getty win on trade marks?
Some versions of Stable Diffusion generated images bearing imitations of Getty Images and iStock watermarks. The court found infringement under sections 10(1) and 10(2) of the Trade Marks Act 1994 for specific examples from older model versions accessed through Stability’s own platforms, but described the findings as historic and extremely limited, and dismissed the reputation claim.
Can IP Global Guard assess our AI exposure in the UK and the EU?
Yes. We review training data, licences, distribution channels and supplier contracts, and coordinate trade mark and copyright strategy across Europe, Latin America and Africa from a single point of contact. Where UK court proceedings are needed, we work with qualified local counsel.
How IP Global Guard can help with AI copyright risk
The Getty appeal and the European cases are moving in parallel, and a model that is safe in one market can be exposed in the next. IP Global Guard, the IP services line of META Channel Corporation Limited, aligns AI, copyright and trade mark decisions across more than 25 jurisdictions in Europe, Latin America and Africa with one strategy and one billing relationship.
Tell us which models you build, buy or distribute, and in which countries. We will map where the training, outputs and distribution create risk and what to fix first. Talk to our AI and IP team.
This article is general information, not legal advice, and reflects the position on its publication date.
Sources
- High Court of England and Wales, Getty Images (US) Inc & Ors v Stability AI Ltd [2025] EWHC 2863 (Ch) (4 November 2025)
- CMS, Getty Images v Stability AI: permission to appeal at the consequentials hearing (21 January 2026)
- CMS, AI and copyright case tracker: Getty Images v Stability AI (updated 16 April 2026)
- Landgericht München I, press release, GEMA v OpenAI, 42 O 14139/24 (11 November 2025)
- AI Act, Recital 106 (2024)
