8 Intellectual property rights

A key legal risk with using GenAI tools is that the datasets used to train the AI models often use and contain copyright material. For example, the internet scraping of online content to train GenAI models can be regarded as an infringing act under UK law.

Maximise

Seven challenges for users of GenAI

Show description|Hide description

A flowchart titled ‘Seven Challenges for Users of GenAI’ which illustrates the potential copyright and third-party rights risks associated with Generative AI. It begins with a central node: Problematic third-party rights for (pure) users of AI. This central issue branches into three main categories: 1. Model – contains content subject to third-party rights. Leads to: Use as such already violates third-party rights. Then to: Use of the model as such already violates third-party rights. 2. Input (Prompt) – contains content subject to third-party rights. Leads to: Use as such already violates third-party rights and Use for intended training violates third-party rights. Because the content was already included in the model. 3. Output – contains content subject to third-party rights. Leads to: Because the content was already included in the input. The content is a purely random parallel creation and the provider itself reserves rights to certain content. The flowchart highlights how third-party rights can be infringed at multiple points in the AI workflow – from model training to prompt design, to output usage – regardless of the user's intent.

Seven challenges for users of GenAI

Current UK copyright law, based on the Copyright, Designs and Patents Act 1988 (CDPA), is ill-equipped to handle the challenges posed by GenAI.

AI developers often use copyrighted content in training datasets without permission, raising infringement concerns. Section 17(2) prohibits reproducing substantial parts of protected works without consent, an issue at the core of ongoing cases like Getty Images v Stability AI [2025] EWHC 38 (Ch).

Authorship under UK law for computer-generated works also remains ambiguous – and this applies to AI-generated outcomes. Section 9(3) of the CDPA 1988 assigns authorship to whoever made the arrangements for creation, but determining this in AI contexts – programmer, prompter, or developer – is complex. The originality requirement further complicates matters, especially for content merely recombined by GenAI.

The UK Government is continuing to explore broadening the exception for text and data mining to support AI research. However, creative industries raised concerns about the lack of technical safeguards for opting out. Broader exceptions could fuel innovation but risk weakening IP protections without enforceable transparency requirements.