In a First-of-its-Kind Ruling, a Delaware Federal Court Rejects an AI Company’s Fair Use Defense to Using Copyrighted Material to Train an AI Model

On February 11, 2025, Third Circuit Judge Stephanos Bibas (sitting by designation) rejected an AI company’s fair use defense on summary judgment in Thomson Reuters Enterprise Centre GMBH v. Ross Intelligence Inc., No. 1:20-CV-613-SB, 2025 WL 458520.

In the other AI copyright cases – with plaintiffs including authors, artists and news outlets alleging their copyrighted material was used to train large-language models such as ChatGPT, Gemini and Llama – the applicability of Thomson Reuters is sure to be hotly debated.

While there are findings in Thomson Reuters that are helpful to those plaintiffs, there are also significant factual differences that the AI developer defendants will highlight in their efforts to distinguish the case.

Background

Ross Intelligence was an AI startup that competed against Westlaw, owned by Thomson Reuters. To train the AI that ran its legal-research search engine, Ross asked Thomson Reuters to license Westlaw’s content, but Thomson Reuters refused.

Ross then gained access to Westlaw’s content through a subcontractor that engaged attorneys to use Westlaw headnotes to create legal questions with good and bad answers. This information was compiled into “Bulk Memos,” which Ross used to train its AI search tool.

In May 2020, Thomson Reuters sued Ross for copyright infringement. This was the first lawsuit alleging that an AI developer trained its model by infringing copyrighted material.

Ross has since gone out of business.

Motions for Summary Judgment Denied, Then Granted

In September 2023, Judge Bibas denied Thomson Reuters’ motions for summary judgment on copyright infringement and the fair use defense, finding issues of material fact.

Last week’s opinion opened with an adage: “A smart man knows when he is right; a wise man knows when he is wrong.” Judge Bibas then reversed course, granting most of Thomson Reuters’ motions for summary judgment on copyright infringement and on the fair use defense.

Ross Failed to Prove Fair Use

The fair use defense involves consideration of four factors: (1) the use’s purpose and character, including whether it is commercial or nonprofit; (2) the copyrighted work’s nature; (3) how much of the work was used and how substantial a part it was relative to the copyrighted work’s whole; and (4) how the infringing use affected the copyrighted work’s value or potential market. The court noted that the fourth factor is most important, followed by the first.

The court found the most significant factors – the fourth and the first – in Thomson Reuters’ favor, and the second and third in Ross’s favor. On balance, the test favored Thomson Reuters.

In rejecting the fair use defense, the court identified the following key considerations:

• Ross was a competitor of Westlaw.
• Ross’s use was not transformative, because it did not have a “further purpose or different character” from Westlaw.
• “Because the AI landscape is changing rapidly, I note for readers that only non-generative AI is before me today.”
• Ross’s argument that it engaged in permissible intermediate copying – e., where the copying is performed as an interim step in a transformative process, and the original work is not shown to the public – failed because the copying was not necessary for innovation.
• As to the most important factor (the fourth): (i) there was a potential derivative market for the copyrighted material: licensing for training of AI models; and (ii) any public benefit was insufficient.

Implications for the Generative AI Copyright Litigation

For the other AI copyright cases, Thomson Reuters’ recognition of a derivative market for the licensing of copyrighted material for AI training purposes is very useful to plaintiffs. As this is the critical fourth factor in the fair use analysis, plaintiffs will contend that the fair use defense fails.

The AI developers will argue that Thomson Reuters is distinguishable in other respects, mainly because the purpose of the developers’ copying was not to compete with the copyright holders, but rather to create the transformative new technology of generative AI, which has substantial public benefits. These are material distinctions, as Judge Bibas himself noted in the opinion. 

In sum, while Thomson Reuters is the first word on fair use in AI copyright litigation, it will not be the last.

Author William Hill is a partner in the Litigation Department at Klehr Harrison.