In a landmark ruling for the artificial intelligence sector, the English High Court delivered a judgment on November 2025 in the case of Getty Images (US) Inc & Others v Stability AI Ltd, addressing crucial copyright issues surrounding generative AI models. This decision marks the first time that UK courts have directly considered whether such AI models infringe copyright when trained on copyrighted material, particularly in the context of the contentious relationship between creative rights and technological advancement.
Getty Images, a leading stock photo agency, initiated the lawsuit against Stability AI, the developer of the open-source generative AI tool Stable Diffusion, which generates images from text prompts. Getty accused Stability of using millions of its copyrighted photographs without permission to train its AI model, utilizing images sourced from a publicly available database called LAION-5B.
Getty’s claims hinged on the assertion that Stability’s actions constituted copyright infringement under the Copyright, Designs and Patents Act 1988 (“CPDA”). However, the court ruled against Getty on copyright grounds while allowing a limited trademark claim. Although the ruling does not have binding authority in Ireland, it provides critical insights for creators, developers, and rights holders regarding AI applications and intellectual property, especially given the parallels between English and Irish copyright law.
The court examined Getty’s trademark infringement allegations under sections of the Trademarks Act 1994 (“TMA”), focusing on claims that Stability’s AI outputs contained watermark-like features resembling Getty’s registered trademarks, which could lead to confusion regarding the source or harm to the brand’s reputation.
Mrs. Justice Joanna Smith assessed each aspect of Getty’s claims. Under section 10(1), which deals with double identity, the court found only minimal infringement, identifying a limited number of iStock watermarks as infringing. The ruling under section 10(2), concerning the likelihood of confusion, revealed that limited infringement occurred, but there was insufficient evidence that later iterations of Stability’s models, SD XL and v1.6, produced infringing watermarks.
Moreover, the court dismissed Getty’s claims under section 10(3), related to trademark dilution, citing a lack of evidence demonstrating any change in economic behavior. Getty also alleged secondary copyright infringement under sections 22, 23, and 27(3) of the CPDA, arguing that Stability’s models amounted to infringing copies because they were developed using copyrighted images without consent.
Stability contended that the training process occurred outside the UK, asserting that UK copyright law did not apply. They further argued that the models do not store Getty’s images but rather contain “model weights” that represent mathematical patterns derived from analyzing the copyright-protected images. The court largely sided with Stability, concluding that these model weights are statistical representations and do not reproduce Getty’s copyrighted works, thus clarifying that AI models trained on copyrighted materials are not infringing copies under current English law.
The court’s decision emphasizes that digital formats, including model weights, can be considered relevant “articles” in the context of secondary infringement provisions. However, it found that the model weights do not store or replicate Getty’s copyrighted images, leading to the dismissal of secondary infringement claims. Mrs. Justice Smith noted that the findings are “historic and extremely limited in scope,” reflecting a judicial preference for evidence-based reasoning rather than broader policy considerations, leaving pivotal questions about AI training and copyright unresolved.
For AI developers, the ruling indicates that training conducted outside the UK may shield them from primary copyright infringement claims, although the judgment does not provide a blanket endorsement for foreign training as a safe harbor. Developers must still navigate potential trademark risks when creating outputs that may unintentionally replicate branding or watermarks, underscoring the importance of conducting due diligence on datasets to mitigate these risks.
Rights holders may face challenges in proving copyright infringement when AI models are trained outside their jurisdiction. Substantial technical evidence will be necessary to trace the use of copyrighted works within AI models. Consequently, rights holders are advised to scrutinize licensing agreements regarding AI training and consider incorporating terms that limit such uses.
Brand owners, while seeing Getty’s trademark claims largely dismissed, are reminded of the persistent legal risks associated with brand misuse in AI outputs. Monitoring generative AI outputs for unauthorized use of logos or trademarks remains essential, necessitating prompt enforcement actions as needed.
The verdict in Getty Images v Stability AI provides significant clarification on technical aspects of AI models and intellectual property laws under English jurisdiction. However, the broader implications of the ruling remain uncertain, as the court refrained from addressing whether the act of training itself constitutes primary copyright infringement. While the decision offers narrow insights into AI and copyright relationships, it leaves many fundamental legal questions surrounding AI training and outputs unresolved.
See also
IMSS Launches Advanced Hemodynamics Unit, Elevates Cardiac Care Amid Ethical AI Governance
Biometrics Market Sees 50 Acquisitions in 2025 as AI Fraud Fuels Urgent Consolidation
CrowdStrike Surges with Strong Q4 Guidance Amid AI Drive and AWS Partnerships; Risks Loom
Arista Networks Eyes Strong Year-End Gains Amid AI Demand and $183B Market Cap
