Summary:
Le 4 novembre 2025, la Haute Cour d’Angleterre et du Pays de Galles a rendu son jugement dans l’affaire Getty Images (US) Inc. & Ors. contre Stability AI Limited [2025] EWHC 2863 (Ch). L’affaire traite de la question de savoir si le modèle d’IA Stable Diffusion constitue une copie contrefaisante d’œuvres protégées par le droit d’auteur en vertu de la Loi sur le droit d’auteur, les dessins et les brevets (CDPA). La Cour a conclu que les poids du modèle Stable Diffusion ne constituent pas des copies contrefaisantes, car ils ne stockent ni ne reproduisent d’œuvres protégées par le droit d’auteur dans leur itération finale ; le fait que le développement ait impliqué une reproduction temporaire ne rend pas le modèle résultant une copie contrefaisante, et le concept juridique de copie contrefaisante requiert la présence effective du matériel protégé par le droit d’auteur à un moment donné.
Original Link:
Generated Article:
The recent judgment in Getty Images (US) Inc. & Ors. v Stability AI Limited [2025] EWHC 2863 (Ch), delivered by Mrs Justice Joanna Smith DBE, highlights a pivotal legal discussion surrounding the application of copyright laws to artificial intelligence (AI) models trained on protected works. This case delves into whether AI systems such as Stability AI’s Stable Diffusion can be considered an infringing copy of copyrighted works under UK copyright law.
At the core of Getty Images’ claim is section 27(3) of the Copyright, Designs and Patents Act 1988 (CDPA), which deals with secondary infringements of copyright. Getty Images argued that the making of Stable Diffusion’s model weights constituted an infringement as they were trained on copyrighted works. However, the court determined that Stable Diffusion does not store or reproduce any copyrighted works within its model weights. Instead, these weights represent mathematical representations derived from patterns and features of the training data. By this reasoning, the court ruled that the AI model, as it exists in its final iteration, cannot be deemed an infringing copy.
The ruling draws heavily from section 17 of the CDPA, which defines the concept of ‘copying’ in copyright law. According to Justice Smith, the concept of an infringing copy requires that the article in question must have at some point contained a copy of the copyrighted material. Stability argued that the nature of AI model weights, which involve intangible, algorithmic interpretations rather than direct reproductions of the content, precludes them from ever constituting an infringing copy. The court agreed and distinguished this case from Sony v Ball, where a RAM chip temporarily stored infringing copies.
From an ethical perspective, the case raises questions about the balancing act between innovation in artificial intelligence and the protection of intellectual property rights. While AI can enable the extraction of patterns and insights from large data sets, including copyrighted content, it is crucial to ensure that creators retain control over their intellectual assets and are not unfairly exploited in the name of technological advancement. The judgment highlights the importance of nuanced understanding of AI technologies when assessing their compliance with existing legal frameworks while encouraging the continuation of transformative innovation.
Industry-wide, this ruling has significant implications for AI developers, creators, and copyright holders. Companies developing generative AI tools may find relief in the judgment, as it appears to protect the legality of training AI models on copyrighted data under certain circumstances. This decision could lead to more relaxed legal interpretations for AI companies, potentially accelerating the development and deployment of similar technologies while maintaining a competitive advantage globally. At the same time, the publishing and creative sectors may see this as a setback in their efforts to contend with the ways AI utilizes their intellectual property without compensation or consent.
As AI continues to evolve, policymakers are urged to provide more explicit guidance and comprehensive regulation around its intersection with copyright law. For instance, requirements for transparency and accountability in AI training processes could help mitigate disputes like this. Alongside such regulations, collaboration between creators and AI developers may foster equitable frameworks that allow creative sectors to benefit from technological innovations while protecting their intellectual property.
This case provides clarity regarding the current legal interpretation of AI models under UK copyright law but also underscores the complexity of balancing legal rights with technological progress, a challenge that will grow more pressing as AI applications expand.