US District Court decision in Thompson Reuters v Ross Intelligence on copyright and AI

On 11 February, the US District Court of Delaware issued its decision in Thompson Reuters Enterprise Centre GmbH & another v Ross Intelligence Inc, revising an earlier summary judgment opinion and order. Unsurprisingly, because the case involves AI, the decision has attracted a lot of attention. 

Facts

Ross made a legal research search engine that uses AI. To train this search tool, Ross required a database of legal questions and answers. It sought a licence from Westlaw (the well-known legal research platform owned by Thompson Reuters). Unsurprisingly, Thompson Reuters refused to grant a licence to allow Ross to build a product that would compete with its own. So Ross obtained its training data – some 25,000 ‘Bulk Memos’ - from LegalEase. These Bulk Memos had been created using Westlaw headnotes. Thompson Reuters therefore sued Ross. 

Decision

The headline points of the decision are: 

• The judge found direct copying by Ross in relation to 2,243 of the headnotes on Westlaw and therefore granted summary judgment in respect of them. 
• All of Ross’s defences (innocent infringement, copyright misuse, merger and scenes à faire) were rejected. 
• Thompson Reuters prevailed on the fair use defence. 

The 2,243 headnotes were those where the Bulk Memo questions were substantially similar to the headnotes in the sense that the similarity was so obvious that no reasonable jury could find otherwise. Copying for the remaining headnotes will be determined at trial. So the summary judgment was partial and not dispositive of the whole case.

Most interesting is the fair use finding. There are may live cases in the US at the moment in which the issue of whether using copyright material to train AI is fair use. As the fair use doctrine can be quite fact specific, a decision in one case may not be determinative of a decision in a subsequent case. So it may be premature to get too excited about this decision. 

Numerically, it was a 2-2 draw between the parties on the four fair use factors. However, Thompson Reuters won the day on the two most important - the first (purpose and character of the use) and fourth (how use affects the copyright work’s value or potential market) and therefore the court found that there was not fair use. 

Comment

Ultimately, Ross was seeking to build a product that would compete with Westlaw and so the decision is not a great surprise – the use in question was broadly the same rather than being ‘transformative’ in a way that would qualify as a fair use. 

Additionally, the AI in this case was not generative AI.  In a case involving generative AI where new content is created, the fair use analysis could follow a different path, particularly in relation to the first and fourth fair use factors.