The UK High Court’s ruling in the landmark case of Getty Images v Stability AI has significant implications for copyright and trademark laws as they relate to Generative Artificial Intelligence (GenAI). The case, which centers on allegations that Stability AI trained its Stable Diffusion model using datasets that included Getty-owned images, underscores critical issues of copyright infringement and territoriality in a rapidly evolving digital landscape.
In the ruling, Getty Images alleged that Stability AI not only trained its AI model using datasets containing Getty-owned or licensed images, but that some generated outputs bore Getty and iStock watermarks. Stability AI countered that the training occurred outside the UK and that the model weights provided for local use did not constitute “copies” under UK law. The court’s decision hinged on the interpretation of primary and secondary copyright infringement and addressed how territorial limits apply to these claims.
The court determined that primary copyright claims against Stability AI failed because the training and development of the AI model occurred outside UK jurisdiction. As a result, the court did not assess whether training an AI on copyrighted images would, in itself, constitute primary infringement under UK law. This highlights a significant hurdle for rights holders, particularly in an era where the actions of AI developers often span multiple jurisdictions.
In its analysis of secondary infringement, the court noted that an “article” involved in secondary liability must embody a copy of the work itself. Stability’s model weights, which encode the learned parameters but do not store or reproduce the visual content of the training images, were not deemed “infringing copies.” Consequently, the importation or distribution of these model weights in the UK did not constitute secondary dealing. The court indicated that the outputs generated by the model should be evaluated separately based on their individual circumstances.
The implications of this ruling extend beyond the UK, particularly to jurisdictions like Hong Kong and Singapore, which have similar legal frameworks for copyright. In Hong Kong, primary infringement arises from actions conducted without a license within its territory, while Singapore’s Copyright Act establishes that infringement occurs when actions are performed within its borders. Thus, primary claims are likely to be difficult to sustain in cases where the training of AI models occurs overseas.
In both jurisdictions, secondary liability is invoked when individuals engage in activities such as importing or dealing with infringing copies. In Hong Kong, the definition of “infringing copy” closely aligns with its UK counterpart. Given the similarities in legal structure, legal analysts anticipate that Hong Kong courts may adopt similar interpretations as seen in Getty v Stability, wherein model weights that do not embody copyrighted visual content are unlikely to be classified as copies.
Singapore’s legal landscape offers some nuances; its Copyright Act provides a non-exhaustive definition of “article,” allowing copyright owners to potentially argue for a broader interpretation. However, any departure from the UK ruling would require litigation to test the merits of such claims. Additionally, Singapore’s statutory computational data analysis exception is broader than that of the UK, potentially allowing for a more permissive approach to copyright infringement findings.
As the digital landscape continues to evolve, the ramifications of the Getty Images v Stability AI ruling are likely to narrow the scope of secondary infringement claims related to model weights and refocus attention on where the actual copying occurs. The emphasis on territoriality means that understanding the specific locations where reproduction or storage takes place will be crucial in future cases.
Looking ahead, both Singapore and Hong Kong face complex legal questions regarding dataset governance, lawful access, output filtering, and licensing in the context of GenAI. As courts grapple with these issues, the balance between protecting copyright owners and fostering innovation among GenAI developers will be increasingly scrutinized. The outcomes of future litigation may shape the very framework within which these technologies operate, influencing the responsibilities and rights of both creators and AI developers moving forward.
See also
OpenAI and Amazon Pursue $10B Investment to Propel AI Chips, Valuation Surpasses $500B
Amazon Plans $10B Investment in OpenAI, Signaling Major Shift in AI Chip Competition
Nvidia CEO Jensen Huang Predicts Gradual AI Job Shift, Not Rapid Replacement
Executives Must Adapt: AI Challenges Traditional Strategy, Demands New Decision-Making Models
Appy Pie Launches AI Video Generation Automation for Streamlined Content Production
