The rise of generative AI has sparked a legal debate on copyright implications.
- Generative AI systems are capable of creating content by learning from existing works.
- The risk of copyright infringement arises during both training and creation phases.
- UK copyright law poses challenges in defending against infringement allegations.
- International standards, such as the EU’s AI Act, may affect future AI development.
The emergence of generative AI systems has introduced significant challenges to copyright law, as these technologies develop new text, images, and other content by learning from existing works. The potential for copyright infringement is most pronounced during the AI training phase, where third-party texts or images might be used to enhance the system’s learning capabilities. While some processes involve data analysis rather than storing copyrighted works, even temporary reproductions risk breaching copyright unless clear permissions, such as public domain or licensing, are in place.
Post-training, AI systems might utilise learned styles to create new content, potentially leading to further copyright concerns. The intricacies involved in identifying specific parts of original works within AI-generated content complicate these issues. UK law, unlike some other jurisdictions such as the US, presents limited defences like fair use, making it challenging to justify instances of alleged copyright infringement.
Relevant UK legislation includes sections 28A and 29A of the Copyright Designs and Patents Act 1988, which allow for transient copies and computational analysis under specific circumstances. However, these exceptions are less applicable to commercial AI projects. Developers seeking to mimic a specific style might explore the defence of pastiche under Section 30A, yet current case law offers limited guidance.
A high-profile case, Getty Images v Stability AI, is poised to set a precedent in the UK regarding such copyright issues. The proceedings could elucidate on primary and secondary infringements and potential defences, with the AI’s training location being a pivotal point. The outcome could offer much-needed clarity on the copyright implications of AI systems.
Looking internationally, the EU Artificial Intelligence Act mandates compliance with copyright laws and promotes transparency by requiring AI developers to disclose training data. The EU’s approach to TDM is more lenient, allowing commercial use unless opted out by the copyright owner. This development implies that AI products from the UK intended for the EU market must adhere to these regulations. Insights from legal experts suggest businesses should establish policies for managing potential copyright risks when deploying AI.
To optimally navigate the complexities of copyright in AI, businesses must stay informed and carefully manage intellectual property risks.
