IP Intelligence

Home | IP Intelligence

Thomson Reuters v. ROSS, 1:20-cv-00613-SB, is the first district court case to address fair use and copyright infringement related to training AI models. Judge Bibas granted summary judgment of no fair use upon a balancing of the fair use factors. The court mostly relied on the findings that (1) ROSS copied headnotes to develop a competing legal research tool to Westlaw and (2) ROSS meant to compete with Westlaw when doing so. The court also rejected ROSS’s “intermediate copying” defense because the subject of ROSS’s copying was not computer code but written words, and the court found ROSS’s copying was not necessary achieve it new purpose.

core/paragraph

While this ruling is significant, it should not be overly generalized to declare AI developers never have a fair use defense to copying of training data. As Judge Bibas points out, this is not a generative AI case. While the analysis may be relevant to generative AI cases like Alter v. OpenAI, it is not determinative.

core/heading

In 2020, Thomson Reuters (Reuters) alleged that ROSS knowingly, intentionally and willfully infringed plaintiff’s copyrights, taking critical features of Westlaw, without permission or compensation, to develop a competing product.

Just before trial was scheduled to begin in August 2024, Third Circuit Judge Stephanos Bibas, sitting by designation in the District of Delaware, made the surprising move to invite the parties to renew their summary judgment motions, nearly a year after first denying them. Reuters and ROSS both renewed their motions for summary judgment on copyright infringement and fair use.

In yet another surprise, Bibas admitted he was wrong and walked back the criticism of Reuters’s arguments included in his earlier summary judgment opinion to now rule that (1) Westlaw’s editorial content is protected by copyright and (2) ROSS’s copying was not fair use.

On Feb. 11, Bibas ruled on the parties’ renewed motions for summary judgment on copyright infringement and fair use. The court granted Reuters’s renewed motions for summary judgment of copyright infringement and no fair use.

The court mostly granted Reuters’s partial motion for summary judgment on copyright infringement. First, the court refined its copyrightability analysis from its earlier decision to conclude Westlaw’s headnotes and the Key Number System meet the minimal threshold for copyrightability because of the creativity required to distill, synthesize or explain part of a legal opinion and organize legal topics. Second, the court found that ROSS infringed 2,243 of the 2,830 headnotes asserted in Reuters’s motion. Bibas left the question of whether ROSS copied the Key Number System and the remaining headnotes (not submitted by Reuters) for trial. Bibas rejected all of ROSS’s defenses to copyright infringement, finding that not one of them “holds water.” &nbsp;

In what may be of more interest to the AI community at large, the court found that ROSS’s intermediate copying of the works to train its AI model was not fair use. Still, Bibas was careful to note “that only non-generative AI is before [him].” Bibas found that the first and fourth factors — transformativeness and the impact of copying on the market — favored Reuters and tipped the overall balance of the four factors to support summary judgment of no fair use.

ROSS’s use was “undoubtedly commercial,” so the court focused its factor one analysis on whether ROSS’s copying of Westlaw’s headnotes was transformative. Informed by the decision in Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, Bibas held that ROSS’s use was not transformative because of the competitive nature of the use. Importantly, the court rejected ROSS’s “intermediate copying” defense. The decision distinguished this case from the intermediate copying cases Sega,&nbsp;Sony and&nbsp;Google v. Oracle,&nbsp;where courts upheld intermediate copying as fair use, because the subject of the copying here was editorial, not functional — like software code — in nature and did not need to be copied to achieve ROSS’s new purpose.

Factor two weighed in ROSS’s favor because the headnotes were “far from the most creative works” because of their factual nature. The court noted that this factor “has rarely played a significant role in the determination of a fair use dispute.”

Factor three weighed in ROSS’s favor because ROSS’s output to an end user does not include Westlaw’s headnotes. Informed by Authors Guild v. Google, Inc., Bibas analyzed the amount and substantiality of what was made accessible to the public rather than focusing on what was used in making a copy.

Where the fair use analysis turned in Reuters’s favor is factor four, characterized as “undoubtedly the single most important element of fair use.” Bibas considered the likely market effect of ROSS’s copying, including both existing and potential derivative markets. In 2023, the court left this factor for the jury out of concern over what best serves the public interest. But now, the court has found that ROSS’s product posed a direct threat to the market for Westlaw and future AI-training data license opportunities, and thus, factor four favors Reuters.

For a deeper dive into Bibas’s order or to see if other courts follow suit, check out BakerHostetler’s case tracker, where we monitor copyright litigations arising out of advanced generative AI in real time, providing case overviews, current statuses and key legal filings. Our coverage of Thomson Reuters v. ROSS is at <a href="https://www.bakerlaw.com/thomson-reuters-v-ross/" target="_blank" rel="noreferrer noopener">https://www.bakerlaw.com/thomson-reuters-v-ross/</a>.

Surprise Move: Judge Walks Back AI Copyright Ruling in Thomson Reuters v. ROSS