The UK High Court dismissed Getty’s remaining copyright claim against Stability AI, concluding that the AI model and its weights do not qualify as “infringing copies” under UK law. The court emphasized that the model was never stored or reproduced in a manner that replicated or preserved any copyrighted material. This decision carries implications for developers deploying AI models trained abroad within the UK, as they may now face reduced exposure to UK copyright law if they do not retain or reproduce protected works.
The ruling centers on the legal definition of “infringing copies” under the Copyright, Designs and Patents Act (CDPA). The court acknowledged that AI model weights - mathematical representations of learned patterns - do not inherently store copyrighted images. While earlier versions of Stable Diffusion produced watermarks resembling Getty’s trademarks, the court determined these were byproducts of the training process, not direct copies of the original works.
The judgment also clarified that an AI model does not constitute an infringing copy merely because it was trained on copyrighted material. The critical distinction lies in whether the model itself contains a copy of the work. Since weights are derived from patterns rather than direct replication, the court concluded that the model does not infringe copyright under current UK law.
Getty’s trademark claims were partially upheld, but only for earlier versions of Stable Diffusion. The court found that outputs from version 1.2 generated watermarks closely resembling Getty’s registered marks, raising concerns about potential trademark confusion. Users might mistakenly associate the generated images with Getty’s brand, prompting the court to assign responsibility to Stability AI for the use of these watermarks.
The decision underscores the importance of proactive monitoring and filtering mechanisms to prevent unintentional trademark infringement. The court emphasized that Stability AI, by making the model available for commercial use, created a commercial link between its output and Getty’s trademarks.
The ruling also highlighted the challenges of proving trademark dilution in AI-related cases. While the court recognized the potential for reputational harm, it found insufficient evidence to support claims of tarnishment. This leaves unresolved questions about the long-term impact of AI-generated outputs on brand identity and market perception.
For developers and businesses, the ruling underscores the need for vigilance. IP Defender offers a cost-effective solution for monitoring trademark databases across 50+ countries, helping brands identify conflicting or confusable marks before disputes escalate. By tracking national registrations, the tool aids in preemptively addressing legal conflicts.
The case mirrors ongoing debates in data privacy law, where model weights are increasingly viewed as potential personal data. While EU regulators argue weights may contain identifiable information, the UK court took a narrower approach, focusing on the absence of stored copies. This divergence highlights the need for businesses to navigate overlapping legal frameworks with care.
As AI evolves, the interplay between copyright, trademark, and data protection laws will remain a critical legal focus. Businesses must adopt robust monitoring strategies to mitigate risks. IP Defender’s continuous surveillance of trademark databases ensures brands are shielded from infringement and conflicts, offering clarity in an increasingly complex legal landscape.
The absence of a clear legal framework for AI-generated content necessitates ongoing vigilance. Businesses must monitor both their use of AI and the potential for third-party claims. As courts navigate the complexities of digital creation, the balance between innovation and intellectual property protection will continue to shape the future of AI development.