U.S. Court Rules In Favour Thomson Reuters (westlaw), Holds That Ross Intelligence Infringed Copyright For AI Training
In what may be the world's first ruling on whether training AI models with copyrighted materials constitutes fair use, a U.S. federal district court has held that such usage amounts to copyright infringement. With ongoing legal battles worldwide, including OpenAI vs. ANI in India, this judgment serves as a strong reference for future similar cases. The United States District Court for...
In what may be the world's first ruling on whether training AI models with copyrighted materials constitutes fair use, a U.S. federal district court has held that such usage amounts to copyright infringement. With ongoing legal battles worldwide, including OpenAI vs. ANI in India, this judgment serves as a strong reference for future similar cases. The United States District Court for the District of Delaware, under Judge Stephanos Bibas, rejected the fair use defense in a case involving the use of copyrighted works to train an AI powered legal research tool. This decision could have significant implications for AI providers and users, emphasizing the copyright risks associated with using third-party materials to train and develop AI tools.
In December 2020, Thomson Reuters, the parent company of the legal research platform Westlaw, filed a lawsuit against Ross Intelligence, an AI startup that had developed an AI-powered legal research tool. Thomson Reuters accused Ross of copyright infringement, alleging that Ross had trained its AI system using Westlaw's headnotes and other editorial content without permission. Thomson Reuters claimed that Ross had initially applied for a license to use Westlaw's content for training purposes, but the request was denied. Instead, Ross had hired a third-party company, LegalEase, to produce bulk legal memos that incorporated Westlaw's headnotes,which were converted into machine learning training data to enable Ross's AI to retrieve relevant judicial opinions for user search queries matching the questions. However, Ross' tool did not employ generative AI. In September 2023, the District Court of Delaware ruled on cross-motions for summary judgment. The court partially granted summary judgment in favor of Thomson Reuters but denied summary judgment on other issues, citing factual disputes. Specifically, the court found unresolved questions regarding whether Westlaw's headnotes were sufficiently original to qualify for copyright protection and various aspects of Ross' fair use defense. Given these circumstances, the court scheduled the trial to commence in August 2024.
The counsel for Ross argued that using Westlaw's headnotes to train its AI tool was transformative and qualified as fair use under copyright law. They asserted that the content generated by ROSS's AI served a distinct purpose and character, making its use permissible and non-infringing. The plaintiff's lawyers had contended that the employment of Westlaw Bulk Memos by ROSS was not fair use and was thus copyright infringement. They believed that in the unauthorized use of Thomson Reuters' copyrighted headnotes, ROSS had illegitimately copied and misappropriated proprietary material in contravention of the intellectual property rights of the publisher.
In deciding the claims of the two parties Judge Stephanos Bibas noted that the headnotes of Thomson Reuters qualify as meeting the minimal threshold of creativity for copyright protection, since they "distill, synthesize, or explain judicial opinions". He held that even literal quotations from opinions qualify for copyright because their selection constitutes editorial judgment. The court held Ross had infringed more than 2,200 of almost 3,000 headnotes by adopting substantially identical language in its Bulk Memo questions. Applying a "substantial similarity" analysis, Judge Bibas compared the headnotes and Bulk Memo questions and found them more similar to the headnotes than to the original judicial opinions. He stressed that a common user would be aware of the similarity, and being both a lawyer and judge, he was best qualified to make that decision. He finally decided that Ross had plagiarized 2,243 headnotes whose wording closely paralleled the Bulk Memo questions instead of the case opinions.
Having established that Ross's application of Bulk Memos constituted copyright infringement, the court proceeded to evaluate whether use of headnotes constituted 'fair use.' The court used four factors of fair use to determine the argument. To begin with, it looked at the character and purpose of the contested material and whether it was adequately transformative. The court ruled that Ross's utilization of Thomson Reuters' headnotes was not transformative in that it performed the same function—enabling legal research—instead of possessing a different character or additional purpose. He recognized that Ross transformed the headnotes into numerical information for processing by AI but determined that its AI search feature still operated similarly to Thomson Reuters' system. He also dismissed Ross's contention that copying was an acceptable intermediate step, distinguishing earlier cases involving computer code from those involving written words. The court recognized that the minimal creativity in Thomson Reuters' headnotes favored Ross but noted that this factor is rarely decisive in fair use cases. It ruled in Ross's favor on the amount and substantiality factor, pointing out that Ross's public-facing products did not include the protected headnotes. However, the market effect factor strongly favored Thomson Reuters, as Ross aimed to create a substitute legal research tool that directly impacted Westlaw's market.
Relying on the observations, the court held that the fair use defense did not apply and granted summary judgment in favour of Thomson Reuters. Justice Bibas ruled partially for summary judgment in favor of Thomson Reuters on the matter of direct copyright infringement, ruling that Ross had illegally copied thousands of headnotes. He also ruled on summary judgment dismissing Ross's fair use defense, finding that Ross's use of the headnotes was not transformative and caused harm to the market. In addition, he rejected Ross's motions for summary judgment on the copyright claims, permitting Thomson Reuters' infringement claims to go forward. Nevertheless, the sole remaining factual issue—whether some of Westlaw's copyrights have expired—will be resolved by a jury.