UK court rules AI training on copyrighted works is not infringing

The High Court has ruled that training artificial intelligence models on copyrighted material does not constitute secondary copyright infringement under UK law, in a decision arising from the Getty Images v Stability AI case.

The case has drawn significant attention from both creative industry groups and technology companies. Getty Images, a provider of licensed images, brought proceedings against Stability AI, the developer of the Stable Diffusion text-to-image model, alleging that the AI system was trained on millions of Getty's images without permission and that the output occasionally contained unauthorised reproductions of Getty's trademarks, including visible watermarks.

Court's findings

The High Court found that the act of training the Stable Diffusion model on copyright works did not result in the production of an infringing copy of those works. The judgment explained that, after training, the AI model did not store any copy of the original protected works, and the model itself did not represent an infringing copy under English law.

Simon Barker, Partner and Head of Intellectual Property at Freeths, commented that the decision "draws a line to say that training an AI model on copyright works, without storing or reproducing those works in the model itself, does not amount to secondary copyright infringement under UK law."

Barker noted that AI developers "can take some comfort from the case that the mere act of training on large datasets will not of itself expose them to liability for copyright infringement in the UK."

He also highlighted that the decision would have a wider impact, stating, "The judgment strikes a balance between protecting the interests of creative industries and enabling technological innovation. It is likely to influence both future litigation and policy debates on AI and intellectual property, not just in the UK but internationally."

Challenges for copyright holders

James Clark, Data Protection, AI and Digital Regulation Partner at Spencer West, observed that the judgment underscores the challenge facing the creative industry: bringing a successful copyright infringement claim related to the training of large AI models.

He explained, "During the training process, the model is not making a copy of the work used to train it, and it does not reproduce that work when prompted for an output by its user."

"Rather, the model 'learns' from the work, in a similar way to the way that you or I might do so. As an expert report quoted in the judgment explains: "Rather than storing their training data, diffusion models learn the statistics of patterns which are associated with certain concepts found in the text labels applied to their training data, i.e. they learn a probability distribution associated with certain concepts."

Clark suggested this aspect of the ruling would cause concern for creative sector organisations while providing encouragement to AI developers.

Trademark infringement claims

The court did find in favour of Getty Images on limited trademark infringement claims, where earlier versions of Stable Diffusion generated images with unauthorised reproductions of the 'iStock' and 'Getty Images' watermarks. Simon Barker clarified that the risk of infringement arises "if AI-generated outputs reproduce protected trade marks… in a way that could confuse people." Each case will depend on the evidence of confusion or association with the relevant trademark.

Unresolved legal uncertainties

Nathan Smith, IP Partner at Katten Muchin Rosenman, noted some uncertainty remains, stating, "On the surface, the long-awaited ruling by the English High Court in the Getty Images Inc v Stability AI case may appear to have provided some clarity on the interplay between AI and IP infringement, but in reality there remains significant uncertainty."

Smith added, "The Court's findings on the more important questions regarding copyright infringement were constrained by jurisdictional limitations, offering little insight on whether training AI models on copyrighted works infringes intellectual property rights." He also noted that the findings relating to unauthorised watermarks were based on specific examples and have "minimal practical impact."

Broader implications

The decision has immediate implications for both AI developers and creative rights holders in the UK, prompting debate about the adequacy of existing intellectual property laws in handling the relationship between AI technologies and copyrighted content. The creative industry is likely to continue seeking alternative legal avenues to protect their works, while technology companies may be emboldened to continue training algorithms on data sets containing protected works, at least in the UK.