Technological progress and the rapid development of artificial intelligence are not only transforming the way technology is used but are also significantly affecting the legal ecosystem. The law, which traditionally evolves more slowly than technology, is now attempting to keep pace with innovations that emerge almost daily.

The adoption of new legislation requires time. As a result, the legal system largely relies on case law and on the creative application of existing legal norms to new and technologically complex situations. For this reason, disputes concerning copyright and artificial intelligence are becoming increasingly significant. Since 2023, when the first such proceedings were initiated, the number of these disputes has risen to 108, according to data compiled by the American law professor Edward Lee. Judicial decisions in these cases do not merely resolve individual disputes. They gradually shape legal frameworks, guide the work of institutions and provide an important foundation for future legislative initiatives in the field of artificial intelligence.

In addition to the positions taken by the District Court in Munich in the case GEMA v. OpenAI, discussed in more detail in our previous article, particular attention has been drawn to the proceedings in Getty Images and Others v. Stability AI Limited, in which the High Court of England and Wales adopted significant positions both with respect to procedural and substantive law.

Getty Images and Others v. Stability AI – Reasoning of the Judgment and the Legal Positions Adopted by the High Court of England and Wales 

Getty Images (US) Inc., Getty Images International U.C., Getty Images (UK) Limited, Getty Images Devco UK Limited, iStockphoto LP and Thomas M. Barwick, Inc. (“Getty and others”) initiated proceedings in the United Kingdom in early 2023 against Stability AI (“Stability”), a company developing generative AI models, in connection with the development, training and deployment of the deep learning model known as Stable Diffusion.

During the proceedings, Getty and others alleged that Stability had, in the course of developing and training the Stable Diffusion model, unlawfully copied millions of visual works from Getty’s websites without authorization. Those works constitute original works protected by copyright, the rights in which are held by Getty and others on the basis of exclusive license agreements concluded with the respective authors. The works used for the development and training of the AI model also contained watermarks incorporating the Getty Images or iStock trademarks, forming the basis for the claim of trademark infringement.

These copyrighted works are particularly suitable for use in artificial intelligence and machine learning systems due to their high quality, detailed descriptions and metadata associated with each work.

Based on these factual allegations, Getty and others initially brought a broad set of claims, which were gradually narrowed during the proceedings. The case was ultimately concluded by a judgment of the High Court of England and Wales in late 2025.

The initial claims sought declarations of:

• primary and secondary copyright infringement,
• database right infringement,
• trademark infringement, and
• passing off.

Primary and Secondary Copyright Infringement and Database Right Infringement

In order to understand the Court’s reasoning in this respect, it is first necessary to consider the concepts of primary and secondary copyright infringement and infringing copies under UK copyright law.

The claim for primary copyright infringement was based on the allegation that the defendant had, in the course of developing and training its AI model, used visual works in which the claimants held copyright pursuant to license agreements with the authors.

In its reasoning, the Court explained that a key element of the Stable Diffusion model is a process which starts with a random noise image. A trained neural network, conditioned by the prompt provided by the user, removes the noise, thereby generating an image that is semantically aligned with the prompt. Consequently, the generation of output images does not require the use of the training data itself, nor does the model store that data, even though much of its functionality is indirectly determined by the data used during training.

As it was not established during the proceedings that the training and development of the Stable Diffusion model had taken place in the United Kingdom, Getty and others withdrew their claim relating to primary copyright infringement based on the use of copyrighted works in the training and development of the AI model.

Furthermore, the prompts allegedly used to generate examples of the disputed output images were subsequently blocked by Stability, which constituted an additional reason for the withdrawal of the claim for primary copyright infringement in relation to output images. Given the close connection between the database right claims and the copyright claims, the database right claims were likewise not pursued.

However, the withdrawal of the claim for primary copyright infringement was primarily related to the lack of evidence establishing jurisdiction in the United Kingdom. For this reason, the same claim was raised in proceedings initiated in 2025 in the United States against Stability concerning the same alleged copyright infringement arising from the development and training of the Stable Diffusion model. It is in those proceedings, which are still at an early stage, that a substantive determination of this issue may be expected. The positions adopted by the UK court will undoubtedly have some influence in that litigation, although the extent of such influence remains to be seen.

The claim for secondary copyright infringement, according to which the claimants argued that the Stable Diffusion AI model constituted an infringing copy because the creation of its model parameters would amount to copyright infringement if it had occurred in the United Kingdom, was likewise unsuccessful.

Acts of secondary copyright infringement relate to subsequent dealings with articles that constitute infringing copies, rather than to the act of reproducing a work itself.

Getty and others argued that Stable Diffusion constituted an infringing copy because it had been imported into the UK, namely by downloading the model in the UK and distributing it in the course of business through the Hugging Face platform.

Copyright infringement exists when a person, without the license of the rights holder, imports into the UK an article which is an infringing copy and which that person knows or has reason to believe is, an infringing copy of the work.

Similarly, infringement arises where, in the course of business, a person possesses, sells, sells or lets for hire, or offers or exposes for sale or hire, exhibits in public or distributes an article that constitutes an infringing copy.

Accordingly, the concept of an infringing copy includes situations where:

• an article constitutes an infringing copy if its making would amount to copyright infringement; or
• an article also constitutes an infringing copy if it has been imported into the United Kingdom and its making in that state would have constituted copyright infringement.

In the context of technological development and copyright law, the Court also addressed the question of the form in which the infringing article must exist in order to constitute secondary copyright infringement. In this regard, the Court held that the infringing article need not be a physical object and may also include intangible digital copies.

However, the Court concluded that an article whose creation involved the use of infringing copies, but which does not itself contain or store those copies, does not constitute an infringing copy.

Consequently, the Court found that the Stable Diffusion model is not an infringing copy. Although the model parameters are modified during training through exposure to copyrighted works, the model itself does not store those works. The model’s parameters therefore do not constitute copies of the works but rather represent patterns and features learned during the training process.

The rejection of the claims for primary and secondary copyright infringement also resulted in the rejection of the claim for additional damages based on the allegation that Stability knew that the model constituted an infringement and had acted deliberately.

Trademark Infringement

The claim for trademark infringement remains, and Getty and others arguing that the ordinary use of the Stable Diffusion model by users in the United Kingdom could, in certain cases, generate synthetic images containing their trademarks.

Getty and others alleged that Stability had committed acts of trademark infringement because images generated by Stable Diffusion contained watermarks identical or similar to the claimants’ iStock and Getty Images trademarks and that such signs were used in connection with the provision of synthetic images.

Although not every image generated by the model contains such signs, evidence provided during the proceedings demonstrated that in certain cases the generation of images through the Stable Diffusion model could indeed result in the appearance of signs identical or similar to the claimants’ trademarks.

Relevant users generated images containing such signs using apparently random prompts. Although it was unclear whether those users were located in the United Kingdom, the Court held that there is no difference between users in UK and users in other countries, namely there was no evidence suggesting that the model functions differently in different jurisdictions.

The Court further found that output images generated within AI model Stable Diffusion containing the disputed signs could be perceived by the average consumer as part of Stability’s commercial communication. The presence of such signs results from the functioning of the model and from the training data over which Stability exercises full control and responsibility. By placing such images on the market, Stability engages in a form of commercial communication with consumers.

The Court therefore held that the use of such signs, provided that they are sufficiently visible and assessed in light of the circumstances of each individual image, could create the impression of a material link in the course of trade between the goods and right holder of the trademark. The average consumer could interpret the sign as an indication that the image originates from the claimants or as an indication of an economic link between Getty and others and Stability, including on the basis that the models were trained on the claimants’ content.

In the Court’s opinion, such use takes place in the course of commercial activity with the aim of obtaining an economic benefit and does not constitute private use, as Stability argued during the proceedings. The Court did not accept Stability’s submission that the use in question constituted private use by consumers based on the assumption that the user exercises full control over the process.

In considering the trademark infringement, the Court concluded that there are sufficiently serious issues of fact to warrant a full trial on the merits.

Passing Off 

By the end of the proceedings, Getty and others continued to pursue their claim for passing off, arguing that images generated by the model containing the claimants’ trademarks could lead to a misrepresentation suggesting that the goods or services originate from, or are commercially connected with the claimants.

Having found that the claim for trademark infringement was arguable, the Court did not proceed to determine the passing off claim. The basis for this decision lay in considerations of procedural economy, as the dispute could be addressed through the trademark claims, while determining passing off would require a more complex factual analysis that would not alter the outcome of the proceedings at that stage. 

The Court’s Conclusion

On the basis of the Court’s reasoning, it may be concluded that there are reasonable grounds for upholding the claim for trademark infringement in relation to the generation of the iStock and Getty Images trademarks in certain versions of the AI model, albeit to a very limited extent. Accordingly, the claim for trademark infringement remains open for further determination at trial on the merits, while the Court did not rule on the claim for passing off.

The claimants withdrew their claims for primary copyright infringement and database right infringement, primarily for procedural reasons, as there was insufficient evidence that the acts of reproduction had taken place within the territory of the UK, which is a prerequisite for the application of UK copyright law, while the claim for secondary copyright infringement was dismissed on the grounds that an AI model which does not store copyrighted works cannot constitute an infringing copy, and therefore there is no basis for an award of additional damages.

Key Legal Positions Adopted in the Proceedings

The proceedings in Getty and others v. Stability represent one of the first major European disputes concerning copyright and generative artificial intelligence. Although the Court did not reach a final determination on all claims, the reasoning of the judgment contains several important legal positions that may influence future disputes.

First, the Court held that AI model parameters do not constitute reproductions of copyrighted works, but rather statistical representations of patterns and characteristics learned during the training process. Consequently, the AI model itself, even when trained on copyrighted works, does not constitute an infringing copy under UK copyright law.

Secondly, the Court confirmed that the object of secondary copyright infringement need not necessarily be a physical item and may also include intangible digital objects, reflecting an effort to adapt existing legal norms to modern digital technologies.

In relation to trademark infringement, the Court held that there were sufficiently serious issues requiring a merit decision, particularly where generated images may contain signs identical or similar to the claimants’ trademarks and may therefore cause confusion among average consumers regarding their commercial origin.

Finally, although the Court did not determine the passing off claim, future disputes may further clarify the extent to which AI developers may be held liable for misleading associations between AI-generated content and particular commercial entities.

Rather than a Conclusion

This case raises important questions concerning the balance between the protection of intellectual property rights, on the one hand, and the development of creative industries and artificial intelligence on the other.

In this case, the UK court adopted significant positions, particularly with regard to the finding that AI model parameters do not constitute reproductions of copyrighted works and that the object of secondary copyright infringement may include intangible digital objects.

At the same time, the Court did not determine the issue of primary copyright infringement in relation to the use of copyrighted works in the development and training of the model, as the claimants withdrew that part of their claim. For this reason, it is expected that the key question concerning the legality of using copyrighted works in the training of AI models, as well as the scope of the fair use doctrine, will be addressed in the proceedings concerning the same factual circumstances currently pending in the United States. The outcome of those proceedings may significantly influence the further development of case law and the establishment of an appropriate balance between copyright protection and the development of artificial intelligence.

This article is to be considered as exclusively informative, with no intention to provide legal advice. If you should need additional information, please contact us directly.