Thomson Reuters Triumphs in Copyright Case Against Ross Intelligence

A U.S. District Court has ruled in favour of Thomson Reuters in its copyright infringement case against Ross Intelligence, a legal AI startup. The ruling, delivered on Tuesday by Judge Stephanos Bibas of the Delaware District Court, is a critical development in the ongoing discussion surrounding AI training and the utilisation of copyrighted materials. Initiated in 2020, this lawsuit is among the pioneering cases that address whether AI firms can engage with copyrighted content without obtaining permission under the “fair use” principle.

Significant Insights from the Court’s Decision

• The court dismissed Ross Intelligence’s fair-use defence, determining that its AI wrongly utilised Thomson Reuters’ exclusive legal content.
• This decision concentrates on non-generative AI, which means it does not yet establish a direct precedent for large language models (LLMs) utilised by firms such as OpenAI and Microsoft.
• Ross Intelligence ceased operations in 2021, citing financial difficulties stemming from ongoing legal challenges.

The copyright lawsuit emerged after Ross Intelligence used Thomson Reuters’ Westlaw database to develop its AI-driven legal research engine. Although legal decisions themselves cannot be copyrighted, the headnotes from Westlaw—summaries of legal rulings authored by human editors—are indeed protected.

Judge Bibas concluded that Ross converted these headnotes into numerical data to train its AI, thus reproducing Thomson Reuters’ work without consent. The court classified this as copyright infringement and rejected Ross’s claim that the data served merely as “added noise” in its AI training.

“None of Ross’s potential defences is viable,” Judge Bibas stated in his ruling. He stressed that Ross had developed a direct competitor using Thomson Reuters’ proprietary content, which significantly affected the market value of the original work.

Implications for the AI Sector

This ruling may have significant repercussions for AI firms, such as OpenAI, Microsoft, and Google, all of which have encountered copyright lawsuits regarding the use of scraped data in training their AI systems. Although this case involved non-generative AI, it prompts broader considerations about whether LLMs trained on copyrighted materials can assert fair use protection.

Currently, a number of similar lawsuits are progressing through the judicial system, and many AI developers are attentively observing how judges interpret the dynamics between AI technology and copyrighted content.