Insights

Key AI case (Getty & Stability AI) flags AI model issues and UK, EU, US Differences

Key AI case (Getty & Stability AI) flags AI model issues and UK, EU, US Differences

Nov 17, 2026
Download PDFDownload PDF
Print
Share

Getty vs Stability AI – UK Court of Appeal Hearing

In a key AI case, the UK Court of Appeal (Getty v Stability AI) heard arguments this past week going to the heart of how AI models are trained and used, and how content owners can (or cannot) use IP rights to sue for alleged infringements by AI companies.

The case is of added interest as it highlights differences of approach by courts in different countries, notably the US v UK v EU, and also how governments are struggling to agree on an approach to statutory updates to keep pace with AI “problems”, all while not wishing to adopt laws that will drive out AI investment in the country.

In the UK case, Getty dropped the primary argument around copyright infringement as the alleged copying of its images took place in the US and so the UK court did not have jurisdiction. The appeal, therefore, concentrated on arguments of secondary infringement and if a model trained overseas on copyrighted images can still be an infringing copy under the statute (CDPA 1988) given the importation and distribution in the UK of the infringing “article”.

A main point of contention was whether the AI model (Stable Diffusion) could be caught if the model weights do not store copies of the original images. Stability AI’s arguments were that it should not because no copies of the underlying works are contained. We await to see if non-storing model weights amount to infringing copies (or, as Stability AI argue, such a model is not an “article” that can be “imported”).

It also flagged again the pressures from different industries, with the music industry being represented by the BPI (British Recorded Music Industry) who intervened in support of Getty, arguing it was dangerous to exempt offshore trained AI models.

The contrast was also made to wins in Munich Regional Court by the German collecting society GEMA against Open AI and Suno, with the German court ruling that training AI on song lyrics (involving “memorization” of works in its statistical parameters) is unlawful reproduction inside the model itself, and even if the training took place in the US the hosting and offering of the memorised model weights in Germany amounted to copyright infringement.

The UK Government had published earlier in 2026 its formal Report on Copyright and AI, confirming these tensions. A proposed “Opt Out” exemption for Text and Data Mining (TDM) was dropped after fears it would encourage uncompensated scraping, while plans to update the CDPA to close offshore trained AI loopholes were shelved, proposals instead focused on building a licensing exchange.

In the US, similar cases have not centered on jurisdictional loopholes, as most AI models were trained using servers in the US. The main battleground has been Fair Use with AI companies arguing the training is highly transformative and also that works are not copied but rather analysed for patterns and statistics etc. The Getty v Stability AI battle continues in the US, being refiled in the Northern District of California. Unlike the UK case, Getty has not dropped its primary copyright infringement claims. Other cases involving Anthropic, Meta, Open AI and others have shown varying views on Fair Use.

Aside from copyright, it is worth flagging that Getty did win in the UK High Court ruling on some (not all) trademark infringement grounds, over the fact Getty Images and iStock watermarks did appear in some generated content (and even though many were not perfect/blurred). Section 10(3) claims for damage to reputation and unfair advantage were dismissed, however, 10(1) and 10(2) claims for identical mark and likelihood of confusion infringement were upheld. Stability AI was denied permission to appeal this aspect and so the findings are final.

Key Takeaways

Key trademark takeaways are that there was no passing the buck to users by the AI provider, the court rejected arguments that use of the mark was the user typing the prompt. Rather, it was the AI provider who was responsible for use of the identical mark “in the course of trade” for the same services (digital images). For the distorted images created by the model, the court view was UK users would still think there was official licensed use.

Key practical takeaways include the fact that AI providers should note the court’s rejection of “mere conduit” type arguments from early ecommerce days. Here, the “mere tool” arguments failed given the outputs of third-party marks.

However, another takeaway was that output filters can be vital. Stability AI did avoid some liability for its updated versions of its model (as it was then using output filters). Finally, depending on where the court finally lands re the article and importation copyright debate, serving models via remote cloud APIs rather than model weight downloads in the UK could help reduce liability for AI output in the UK.

Looking at Getty v Stability AI as a whole and general themes impacting generative AI cases, it is clear that each case is very fact and evidence specific.

The UK Court of Appeal written decision in Getty v Stability AI is expected sometime in spring/summer 2027.

Whatever the outcome, the losing party could well seek permission to appeal to the Supreme Court, so we may well have some more twists and turns to come. Watch this space!

Related capabilities

  • Digital Transformation & Emerging Technology

  • Data Privacy & Security

Meet the team

This material is not comprehensive, is for informational purposes only, and is not legal advice. Your use or receipt of this material does not create an attorney-client relationship between us. If you require legal advice, you should consult an attorney regarding your particular circumstances. The choice of a lawyer is an important decision and should not be based solely upon advertisements. This material may be “Attorney Advertising” under the ethics and professional rules of certain jurisdictions. For advertising purposes, St. Louis, Missouri, is designated BCLP’s principal office and Kathrine Dixon (kathrine.dixon@bclplaw.com) as the responsible attorney.