What does the Thomson Reuters Ross Intelligence decision mean for UK AI companies?

‘This decision may not be the slam dunk rightsholders were hoping for’

With legal precedent on AI copyright starting to be set the implications will be felt around the world, says Broadfield Law solicitor Tancred Campbell.

There is a raft of important court decisions expected in 2025 that will set the stage for future AI development. The first was delivered in February in a claim brought by Thomson Reuters against Ross Intelligence Inc.

The case provides an early indicator as to how claims brought by copyright owners against developers of AI tools may be decided, and the implications for AI companies operating from the UK.

Thomson Reuters owns a widely used legal research platform called Westlaw, and with it the copyright over the content on the platform. Ross Intelligence built a legal AI search tool using the ‘headnotes’ that appear on the Westlaw platform as training materials. The alleged copying in this case was relatively clearcut, as Thomson Reuters was able to show the answers the Ross Intelligence model generated substantially copied those headnotes.

It was a case where the concept of ‘fair use’ that exists under US copyright law was used as a defence.

Copying copyrighted materials in the US may be considered ‘fair use’ if it can be shown that the copying was, for example, for the purpose of criticism, comment, news reporting, teaching, scholarship or research.

Other factors need also be considered, such as whether use is of a commercial nature, the nature of the copyrighted work, the amount of copyrighted work used and the effect on the market value of that copyrighted work.

In this instance, the judge sided with Thomson Reuters - placing particular emphasis on the fact that Ross Intelligence had copied the Westlaw headnotes to build a tool that could compete with Westlaw commercially.

Implication for UK AI developers

For developers in the UK, the decision against Ross Intelligence should send a clear warning that the fair use defence under US copyright law is unlikely to offer any salvation where the model being trained could conceivably compete with the product or services of the copyright owner.

Nonetheless, this decision may still not be the slam dunk that rightsholders were hoping for.

The Ross Intelligence model was not generative; rather, when prompted, it only provided pre-prepared answers that were almost identical to the training materials.

The US judge took pains to emphasise this point, which leaves open the question whether copyright infringement can be said to take place when a model is trained on copyrighted training materials but the output materials it generates are not sufficiently similar to the materials it was trained on to meet the threshold for copyright infringement.

There are substantial similarities between copyright law in the UK and US. In the UK, developers may look to rely on a ‘fair dealing’ defence for text and data mining for research purposes. However, a clear distinction exists under UK law: fair dealing must be ‘for a non-commercial purpose’ only.

Though we await the decision in several key UK cases, it seems a safe bet to surmise that a literal reading of the Copyright, Designs and Patents Act 1988 would exclude any data and text mining of copyrighted materials carried out for any commercial end.

The Ross Intelligence decision placed significant emphasis on its tool being a competitor of Westlaw - to the extent that, if its tool was not a competitor, the fair use defence may have succeeded. It appears therefore that copyright law in the UK is, at this moment in time, less favourable for developers, as it excludes data and text mining for commercial purposes full stop.

Copyright law clarification

Copyright law in the UK has been ripe for clarification ever since it became apparent that AI tools needed reams of data for training purposes. This is something the government is proposing to address, having launched an open consultation on copyright and AI.

The government’s stance is that creating a legal environment that allows developers to build world-leading tools is crucial to economic growth, though it does seek to acknowledge those individuals and business that rely on copyright to protect their interests.

The government has put forward a preferred approach: namely, to codify an exception to copying infringement for text and data mining of any copyrighted materials that can be accessed by lawful means (i.e. available online or through a subscription), including text and data mining for commercial purposes.

Rightsholders would be able to ‘reserve’ their rights through an ‘opt out’ so that the exception would be disapplied to their work, but the default position would be that text and data mining would not amount to copyright infringement.

Though still at the consultation stage, it would certainly come as a surprise if the government were to substantially move from its preferred approach. Rightsholders have, unsurprisingly, voiced their concern that the government’s preferred approach would place developers’ interests above those of the creators of the works.

On balance, the government’s proposed exception would likely create a more favourable legislative framework than that which currently exists. However, much will depend on how the rights reservation mechanism will function in practice, and how strict the requirements on developers to demonstrate their compliance with the new framework will be.

The government’s proposed changes are timely, as clarity under copyright law is required. However, it is also possible that the government has taken the view that copyright law currently prohibits the training of generative tools on copyrighted materials. If this was the case, then the proposed legal changes may be intended to address the stifling consequences for economic growth that may flow from a decision against the developers of generative AI tools in the verdicts we expect later this year.