Thomson Reuters v. Ross: The AI-Training Fair-Use Decision

A 2025 ruling rejected fair use for copying Westlaw headnotes to build a competing research system.

What happened—and what the record establishes

Thomson Reuters alleged that Ross Intelligence used copyrighted Westlaw headnotes to help train a competing legal-research tool. In a 2025 summary-judgment opinion, the court found infringement as to thousands of headnotes and rejected Ross’s fair-use defense for the uses before it.

The court emphasized commercial competition and concluded that the use was not transformative. It also noted that Ross’s system was not generative AI. That distinction matters: the opinion is influential, but it did not decide every question about training modern generative models on books, images, music, or web pages.

Fair use is a fact-intensive, multi-factor defense. Purpose, nature of the works, amount used, market effect, access, copying, and output behavior can produce different results in other cases.

Questions the law must answer

• Were protectable expressive elements copied?
• Did the secondary use serve a different purpose?
• Did the use substitute for an existing or reasonably likely licensing market?

Practical takeaways

• Training is not automatically fair use or automatically infringing.
• The opinion concerned a non-generative competing legal product.
• Businesses should document dataset provenance and licensing decisions.