A Model Decision? The High Court’s Take on AI and IP

The ruling in Getty Images v Stability AI, handed down on 4 November 2025, addresses the fundamental tension between the interests of AI developers and innovators, a balance noted by the Court as being “of very real societal importance” (para 12 of the judgment). While the High Court found for Getty Images on limited trade mark infringement claims, the dismissal of the secondary copyright claim largely benefited AI developers, leading to an outcome that specialists described as a "narrow win" for Getty and a “damp squib” for broad legal clarity.

It should be noted that:

·   An appeal is expected, suggesting the judgment is unlikely to be the final word on the matter

·   The case has been widely described as having a limited outcome, particularly concerning the core issues surrounding AI training and copyright

·   The judgment left many questions unanswered regarding copyright and AI training, strongly suggesting that further legal and political challenges are inevitable as the AI sector expands.

I’m not legally trained and know little about AI – what do I need to know about this judgment?

The High Court dismissed the crucial copyright infringement claim by Getty Images against Stability AI, ruling that the AI model itself is not an “infringing copy” of Getty Images’ copyright-protected works under UK law, because the model only contains the patterns and features learnt from the training data and does not store or reproduce the original copyrighted works owned by Getty Images.

However, Getty achieved a narrow victory on trade mark infringement, as specific earlier model versions of Stable Diffusion were found to infringe by generating synthetic images bearing signs identical or similar to Getty Images and/or iStock watermarks (though these successful findings were deemed historic and extremely limited in scope).

Why is the judgment significant?

The UK’s approach to AI regulation

The UK's stance on AI regulation is a “pro-innovation” sector-specific and principles-based approach. Instead of new, overarching legislation (like the EU AI Act), it relies on existing regulators to interpret and apply five core principles within their specific domains using existing law and supplementary guidance. The lack of regulation presents challenges for developers, businesses, and lawyers, and critics argue that the UK’s light-touch model may fail to address serious risks, such as infringement, data leakage, and misuse of generative AI.

The key takeaway of the case, favouring AI developers, was the dismissal of Getty Images' secondary copyright infringement claim.

However, the finding of trade mark infringement by Stability AI serves as a warning that if AI-generated outputs confuse consumers about the origin or licensing of content through visible branding artifacts (like watermarks), the developer may still face liability.

What were the key issues explored in Court?

Secondary copyright infringement

Copyright arises automatically, subject to certain criteria being met. Copyright protects things like images, and the law rewards creative individuals/businesses by giving them exclusive rights to use their work and to generate income from it.

Secondary copyright infringement occurs when a party helps or enables another party to infringe copyright, even if they didn’t directly copy the work themselves. It usually involves the third party distributing, selling, or facilitating access to infringing content, and the copyright owner is entitled to seek damages.

Getty Images argued that Stable Diffusion was an infringing copy of their copyright-protected work because its making (the training process) was conducted outside the UK would have constituted copyright infringement if carried out in the UK, as unauthorised storage of the works occurred during training. The dismissal of the claim centred on the statutory construction of the terms “article” and “infringing copy” in relation to Stability AI’s Stable Diffusion model.

Despite the broad interpretation of “article” (an AI system or other software distributed over the internet in intangible form) under the Act, the Judge found that Stable Diffusion was not an “infringing copy” of copyright-protected works under section 27 of the Copyright Designs and Patents Act 1988. The Judge explored the following:

·   An article only qualifies as an “infringing copy” if it constituted an infringement. Critically, to be an infringing copy must, at some point, have stored or contained a reproduction of the copyright work

·   The Stable Diffusion model itself (specifically, its model weights – the learnable parameters that define the network connections in the AI model, which control the functionality of the network) does not store or reproduce any copyright works

·   While the process of training the model involved the reproduction and storage of Getty Images’ content (the act that would have been primary infringement if undertaken in the UK), the final AI model only contains the patterns and features learnt from the training data, not copies of the words themselves.

Therefore, an AI model that results from a training process involving infringing copies, but which never contains or stores those copies itself, is not an infringing copy. The secondary acts to which Getty Images objected fell outside the policy and object of the Act.

Trade mark infringement

Trade mark infringement happens when a party uses a sign that is identical or confusingly similar to a registered trade mark without permission, in a way that could mislead consumers or damage the brand.

Getty Images secured a limited victory through trade mark infringement law, offering a specific way to challenge AI outputs that imitate their brands. The central allegation here was that Stability AI “scraped” millions of Visual Assets (including high-quality photos) from Getty Images’ websites without consent and unlawfully used these images to train and develop various versions of Stable Diffusion. Visual Assets featured on the Getty Images websites display watermarks containing either “Getty Images” or “iStock”, overlaid on the image, often within a grey translucent banner. These watermarks are iconic and must be properly licensed to be removed from an image.

The specific basis for the trade mark infringement claim arose because, after training on this scraped and Getty-watermarked data, the Stable Diffusion model would, in some instances, generate synthetic images bearing Getty Images’ and/or iStock trade marks, specifically in the form of trade marks! Getty Images alleged that Stability AI infringed these marks by affixing signs (“watermarks”) that were identical or similar to the Marks to the synthetic images generated by Stable Diffusion without their consent.

In summary:

·   UK users generating watermarked images: The trade mark infringement claim focused on the distribution and use of the Stable Diffusion model, particularly the fact that normal use by users in the UK could lead to the generation of these watermarked synthetic images

·   Trade mark infringement success: The court found that Stability AI had committed limited infringements under sections 10(1) and 10(2) of the Trade Marks Act 1994, specifically by generating synthetic images bearing Getty Images and iStock watermarks in certain specific and limited instances, particularly involving older models of the AI model

·   Warning to AI providers: This finding acts as a warning that AI-generated outputs that reproduce protected trade marks in a way that creates a likelihood of confusion or association could lead to trade mark infringement liability

·   Rejection of tarnishment/dilution: However, Getty Images’ broader claim under section 10(3) of the Act (detriment to repute, dilution, or unfair advantage) was dismissed due to a lack of evidence of widespread infringement and failure to prove or deduce a change in consumers' economic behaviour.

Unclear legal waters?

The dismissal of the secondary copyright claim led to expert comments that the ruling left the UK legal waters of copyright and AI training as uncertain as before the judgment.

·   Content Creators' Demand for Regulation: Getty Images expressed that well-resourced companies face significant challenges in protecting their creative works, given the current lack of transparency requirements. They called on the UK government to put in place new AI transparency rules to avoid continued exploitation. Under-resourced companies/startups may struggle to enforce their rights, which is problematic

·   Shift to Negotiation: In the absence of clear legal guidance, some observers suggest the outcome may encourage both AI developers and creative industry stakeholders to favour negotiated commercial solutions over prolonged and expensive litigation. For more information about alternative dispute resolution, please see here.

Summary

In summary, the ruling significantly enhanced the legal security for AI developers concerning the training and development of AI models (provided the underlying copyrighted material is not stored or permanently reproduced in the model). Conversely, creatives failed in their primary fight to halt AI training through copyright secondary infringement claims, leaving them to rely narrowly on trade mark law to police brand imitation in AI outputs and pressuring them to seek legislative intervention or commercial licensing agreements to protect their core assets.

Contact our legal experts