Getty Images v. Stability AI: a Landmark Ruling on the Liability of Generative AI Systems for Intellectual Property Infringement

I. Background of the Case

On 4 November 2025, the High Court of Justice in London delivered its long-awaited decision in Getty Images (US) Inc. and others v. Stability AI Limited¹, a landmark case at the intersection of intellectual property law and artificial intelligence.

Getty Images, one of the world’s leading visual content providers, sued Stability AI, the company behind the Stable Diffusion image generation model, alleging the unlawful use of millions of copyrighted photographs from Getty’s databases to train the AI system.

The claimants alleged multiple forms of infringement:

• Primary and secondary copyright infringement under the Copyright, Designs and Patents Act 1988 (CDPA);
• Trade mark infringement under the Trade Marks Act 1994 (TMA), due to the generation of synthetic images displaying Getty’s watermarks (“GETTY IMAGES” or “ISTOCK”); and
• Passing off, on the ground of the misappropriation of Getty’s reputation and goodwill.

The allegedly copied material derived from the LAION-5B dataset, a massive collection of web-scraped image–text pairs assembled by a German non-profit association, LAION e.V. Stability AI admitted that “some” Getty-sourced images were included in the training data but denied any direct liability, arguing that:

• the training took place outside the United Kingdom;
• the AI model itself is not a “copy” within the meaning of copyright law; and
• any reproduction of Getty watermarks resulted from user-initiated prompts, not from Stability’s own conduct.

‍

II. Legal Issues Before the Court

The proceedings raised novel questions on how existing UK copyright and trade mark law apply to machine-learning models that process human-authored content.

‍

1. The legal responsibility for Stable Diffusion

The first issue was whether Stability AI bore legal responsibility for the development, release, and operation of Stable Diffusion versions 1.x and 2.x.

Justice Smith held that the early versions of the model had been published on CompVis GitHub and Hugging Face by academic researchers from Heidelberg University, not by Stability AI.

As a result, the Court found that Stability AI was not legally responsible for the open-source publication of those versions and could not be treated as the principal infringer in respect of the initial releases.

‍

2. Alleged trade mark infringement

Getty Images further alleged that the generation of synthetic images bearing GETTY IMAGES or ISTOCK watermarks infringed its registered trade marks.

The Court undertook a detailed analysis of all three forms of infringement under the Trade Marks Act 1994:

• Section 10(1) – identical mark for identical goods;
• Section 10(2) – similar mark creating a likelihood of confusion; and
• Section 10(3) – unfair advantage or detriment to the reputation of a mark with a reputation.

After considering expert and experimental evidence from both sides, the judge found that:

• certain iStock watermarks reproduced by Stable Diffusion v1.x (notably via DreamStudio and the Developer Platform) amounted to Section 10(1) infringement;
• a small number of cases could also fall under Section 10(2) due to a potential risk of confusion;
• however, no infringement was established under Section 10(3), as Getty failed to prove any reputational harm or unfair advantage.

The Court stressed that these incidents were isolated, historic, and technically corrected in later versions (v2.x and SD XL).

‍

3. The “secondary infringement” claim

Getty Images also pursued a secondary infringement claim under sections 22–27 CDPA, arguing that the Stable Diffusion model itself constituted an “infringing copy” of protected works.

The Court firmly rejected this argument, holding that:

• the trained model does not qualify as an “article” or “infringing copy” under the CDPA;
• it does not store any of the training data; and
• in any event, the training activities occurred outside UK territory, depriving the British courts of jurisdiction over the alleged acts of primary infringement.

Accordingly, the secondary infringement claim failed in its entirety.

‍

4. Copyright ownership and licensing

A further issue concerned the exclusive licensing agreements between Getty Images and its photographers.

The Court confirmed that several template agreements constituted exclusive licences under section 92 CDPA, granting Getty standing to sue.

Nevertheless, this finding had no practical impact on the outcome, since the alleged copyright infringements were not established.

‍

III. The Court’s Findings

After extensive hearings and technical evidence, the High Court rendered a comprehensive 200-page judgment, concluding that:

1. The secondary copyright infringement claim is dismissed. Stable Diffusion is not an infringing copy and falls outside the scope of the CDPA.
2. The trade mark claim succeeds in part. A handful of isolated iStock watermark reproductions amounted to limited infringements under sections 10(1) and 10(2) TMA.
3. The passing-off claim is rejected. No misrepresentation or damage to goodwill was proven.
4. No injunctive or monetary relief is granted. The Court considered the infringements “historic and extremely limited in scope”.

As Mrs Justice Smith summarised:

“Although Getty Images succeed (in part) in their Trade Mark Infringement Claim, my findings are both historic and extremely limited in scope. The Secondary Infringement Claim fails.”

‍

IV. No Financial Sanctions at This Stage

Importantly, this judgment was delivered following a liability-only trial, as ordered by Master McQuail on 22 April 2024 (“a first trial to determine liability”).

Consequently:

• no financial sanctions or damages were imposed;
• the Court expressly reserved all questions of quantum, injunctions, publication orders or further remedies for a potential second-phase trial.

Such bifurcation is common in complex IP litigation, allowing the Court to determine the existence of liability before addressing the quantification of damages.

Thus, while Getty Images achieved a narrow declaratory success on trade mark infringement, no monetary award or injunction was issued at this stage of proceedings.

‍

V. Broader Implications

This ruling is the first in the United Kingdom to address, in depth, the legality of training and deploying generative AI models from an IP perspective.

Its key takeaways are clear:

• The Court confirmed that an AI model is not itself a “copy” of the works used for training, but the result of a statistical learning process.
• The territorial limits of UK copyright law remain decisive: acts of training conducted abroad fall outside domestic jurisdiction.
• The developer of an AI system is not automatically liable for all user-generated outputs; liability arises only from commercial or controlled uses attributable to the developer.
• Trade mark law applies narrowly, capturing only clear instances of confusion or misrepresentation, not incidental artefacts of model training.

This pragmatic approach seeks to balance creative-industry protection with technological innovation, reflecting judicial caution in an area where statutory reform is still evolving.

‍

VI. Conclusion

The decision in Getty Images v. Stability AI stands as a milestone in European and UK jurisprudence on artificial intelligence.

It reaffirms the primacy of territorial and statutory boundaries while signalling the judiciary’s readiness to adapt traditional IP principles to new technological realities.

Future proceedings, notably a possible quantum phase, will determine whether Getty Images pursues financial compensation for the limited trade mark infringements acknowledged.

For now, the High Court has provided the first clear judicial guidance on how generative AI models interface with copyright and trade mark law, setting an influential precedent for subsequent cases across Europe and beyond.