High Court delivers landmark judgment on generative AI and copyright

On 4 November 2025, the High Court delivered its much-anticipated judgment in Getty Images (US) Inc. v Stability AI Ltd. This concluded the UK’s first full trial to consider how copyright and trade mark law apply to generative Artificial Intelligence (AI) model training and outputs.

After Getty withdrew its primary copyright and database right claims earlier in the year, the Court dismissed the remaining allegations of secondary copyright infringement and found only a very limited number of instances of trade mark infringement.

The decision represents a decisive outcome in Stability AI’s favour and offers an early indication of how English courts may approach future disputes concerning AI training data, model weights and generated outputs.

Background - narrowing of the dispute

As we have previously examined, Getty initially alleged that Stability AI had unlawfully used millions of Getty-owned and licenses images to train its Stable Diffusion model, infringing its copyright, database and trade mark rights.

By the time of trial, Getty withdrew its core claims for direct copyright and database right infringement, leaving two questions for determination:

1. Secondary copyright infringement - whether distributing Stable Diffusion model weights via platforms such as Hugging Face amounted to importing or dealing in an “infringing article” under sections 22-23 of the Copyright, Designs and Patents Act 1988 (CDPA); and
2. Trade mark infringement - whether certain Stable Diffusion outputs displaying watermark-like features infringed the Getty or iStock registered trade marks.

Secondary copyright infringement - model weights not “infringing articles”

The Court held that the Stable Diffusion model weights were not “infringing copies” under section 27(3) CDPA and, therefore, could not give rise to secondary infringement.

While the judge accepted that “articles” can include intangible electronic files, she found that an infringing copy must at some point embody, contain or store a protected work. The evidence showed that the model weights did not store any Getty images but instead represented statistical relationships learned during training.

Getty’s argument - that the making of the model would have constituted infringement if carried out in the UK - was rejected. Hypothetical infringement alone was insufficient. The ruling draws a clear line between using copyright material to train an AI model and reproducing that material in a form capable of infringing.

Trade mark infringement - narrow and exceptional findings

Getty also alleged that certain Stable Diffusion outputs displayed versions of its Getty and iStock watermarks, amounting to trade mark infringement under sections 10(1), 10(2) and 10(3) of the Trade Marks Act 1994.

After reviewing hundreds of examples submitted by both parties, the Court found no evidence that UK users typically used Getty-style captions as prompts, or that watermark-like features appeared with any frequency.

Ultimately, the judge identified only three instances - two under the iStock mark and one under the Getty mark - where early versions of Stable Diffusion (v1 and v2) produced images close enough to infringe under sections 10(1) or 10(2). The broader claim under section 10(3) was dismissed entirely for lack of evidence of reputational or economic harm.

The Court rejected Stability’s argument that end-users were responsible for the infringing outputs. It held that, in this case, any infringement arose from the model’s design and training, and therefore responsibility rested with Stability as the model provider controlling the system and its data.

Passing off, title and other findings

The Court declined to make separate findings on passing off, considering that it added nothing beyond the trade mark analysis. It also addressed several procedural issues:

• Title and licensing: The Court found that Getty held exclusive rights under some, but not all, of its sample contributor agreements, underscoring the importance of clear and properly drafted licences.
• Model release responsibility: Stability was not liable for the initial release of Stable Diffusion v1 by the independent CompVis research group, having not authorised or directed those acts. This clarifies that companies are not automatically liable for actions of independent collaborators - only for releases or conduct they authorise or control. For organisations developing or deploying open-source or jointly developed AI models, clear documentation of responsibility is essential to manage downstream legal risk.
• Quantifying training data: The judge declined to make any finding on how many Getty works were used in training, noting that the evidence was insufficient to establish the scale of use. This reflects the broader evidential challenge of proving how specific works feature within large, publicly sourced AI training datasets.

Key lessons from this judgment

This is the first UK judgment to analyse in depth how existing copyright and trade mark law applies to generative AI. While the Court stopped short of redefining Intellectual Property (IP) law for the AI era, several key lessons emerge:

1. AI model weights are not “copies” of training data. Statistical representations, without embedded image files, are unlikely to constitute reproduction.
2. Evidence is paramount. Getty’s difficulties in proving that infringing activity occurred in, or targeted the UK, highlight the challenges of applying territorial copyright principles to global AI development.
3. Trade mark claims face a high evidential bar. Hypothetical or engineered examples will rarely suffice without proof of real-world confusion or harm.
4. Pressure for reform will continue. The case leaves unresolved policy questions around text-and-data-mining exceptions, dataset transparency and how to balance innovation with rights protection.

Implications for rights holders, AI developers and businesses

1. For rights holders: the judgment underscores the importance of provenance tracking and watermark management, while illustrating the evidential difficulty of tracing AI training activity.
2. For AI developers: the ruling offers reassurance that using publicly accessible data for model training - without embedding copies - may not automatically breach UK copyright law. It also confirms that, where infringement does occur, responsibility may rest with the model provider rather than end-users, reinforcing the need for robust governance, filtering and contractual risk management. The case further highlights the importance of defining responsibility in collaborative or open-source development, ensuring that liability is limited to acts within a company’s control.
3. For businesses and deployers: the decision reinforces the need to manage downstream risk through contractual indemnities, due diligence on AI suppliers and awareness of evolving global regulation.
4. For organisations across all sectors: this judgment should serve as a prompt to review or establish internal AI policies to govern data use, model deployment, and employee interaction with generative tools - ensuring alignment with evolving legal, ethical and regulatory expectations.

What next for UK law and AI model training?

While Getty may consider an appeal, the decision marks a defining moment in the legal treatment of AI model training. It establishes that, under current UK law, model weights and statistical outputs fall outside traditional notions of “copying”, leaving broader reform to Parliament rather than the courts.

The ruling also adds momentum to policy work under the Data (Use and Access) Act 2025, with a government report on AI and copyright due in early 2026.

Getty Images v Stability AI is one of the first full-scale UK High Court cases to test how existing IP laws apply to generative AI training and model use. While the judgment provides important clarification generally in favour of AI developers, it also highlights the limitation of the current law and the need for continued transparency and legislative review in this rapidly evolving field.

Hamlins Commercial and Tech team support clients across multiple industries, sharing commercial and regulatory expertise and advising on the protection of intellectual property rights and AI use. If you would like a conversation about how we can help you navigate changes impacting your business, please get in touch.