By: Elizabeth S. Dipchand and Zach Nickels
At a time of considerable uncertainty surrounding the exponential intersection of copyright and AI development, a landmark copyright decision coming from the United States District Court for the District of Delaware represents a significant win for copyright owners, and reinforces the significant control that copyright owners have over their copyright protected works.
On February 11, 2025, the Court granted summary judgment in favour of the Plaintiff Thomson Reuters in the first fair use case involving non-generative artificial intelligence (“AI”), revising a previous 2023 summary judgment opinion and order in the case.
The potential for this decision to set a trend makes it imperative for AI developers to carefully consider how they engage copyright owners and their works in order to avoid liability. This will likely include comprehensive and clear licence arrangements explicitly covering use in training models; alternatively, risks can be significantly mitigated by relying solely on works in the public domain. However, the question as to whether a fair use defence analysis would result differently in the context of generative AI, wherein the impugned use is found to be transformative in nature, still remains unclear.
Reconsideration of Summary Judgement Order
In his decision, Circuit Judge Stephanos Bibas noted that he had previously largely denied Thomson Reuters’ motions for summary judgment on copyright infringement and fair use defence, but in the lead up to the August 2024 trial date revisited the case materials and realized that the prior summary judgment ruling had “not gone far enough”. Judge Bibas invited the parties to renew their summary judgment briefings, and in his decision:
- Granted most of Thomson Reuters’ motion for partial summary judgment on direct copyright infringement and related defenses;
- Granted Thomson Reuters’ motion for partial summary judgment on fair use;
- Denied Ross Intelligence’s motion for summary judgment on fair use; and
- Denied Ross Intelligence’s motion for summary judgment on Thomson Reuters’ copyright claims
Working Around Westlaw
Thomson Reuters Enterprise Centre GMBH (“Thomson Reuters”) owns Westlaw, one of the largest legal-research platforms commonly used by lawyers and law students. For the payment of subscription fees, Westlaw provides its users access to case law, statutes, law journals and treatises among other content, all of which is organized using a “Key Number System”. Westlaw also provides editorial content such as headnotes (“Headnotes”) which summarize the case law and key findings. Thomson Reuters owns the copyright in Westlaw’s copyright protected material.
Ross Intelligence Inc. (“Ross”) is a new competitor on the market, which offers a legal-research search engine that uses AI. In the events leading up to this dispute, Ross needed a database of legal questions and answers to “train” its AI search tool and asked to license Westlaw’s content. Thomson Reuters refused this request because Ross was a competitor.
To circumvent this roadblock and train its AI, Ross entered into a deal with LegalEase to access training data in the form of “Bulk Memos”, which were lawyers’ compilations of legal questions with good and bad answers. LegalEase provided the lawyers that created the Bulk Memos with a guide on how to create the questions using Westlaw’s Headnotes. In essence, Ross trained its AI on the Bulk Memos, which were built themselves from Thomson Reuters’ Headnotes. As such, the Court described the dispute as boiling down to whether the LegalEase Bulk Memo questions copied the Headnotes, or were instead taken from uncopyrightable judicial opinions.
At Issue… Again
- Direct Copyright Infringement
In assessing whether Ross directly infringed Thomson Reuters’ copyrights, the Court considered whether Thomson Reuters owned a valid copyright, and whether Ross copied protectable elements of the copyrighted works. The second element required the Court to determine whether Thomson Reuters had shown that Ross actually copied the works, and that the copying was substantially similar to the works at issue.
Thomson Reuters’ Copyrights are Original and Valid
With respect to the validity of Thomson Reuters’ copyrights, the Court recognized that originality is central to copyright and a low threshold to meet, only requiring some minimal degree of creativity. The pertinent question is whether a work is original, not how much effort went into creating it. The Court found that the Headnotes and Key Number System met the minimal threshold for originality, satisfying the first element.
Ross Copied Protectable Elements of the Copyrighted Works
With respect to whether Ross copied protectable elements of the copyrighted works, the Court considered whether there was copying of the constituent elements of the works that were original, deciding whether Thomson Reuters had proven both actual copying and substantial similarity. The Court applied the actual copying and substantial similarity analyses to a subset of 2,830 Headnotes and found actual copying of 2,243 of them. However, the Court reached no new decision on the Key Number System as factual disputes remained as to whether Ross had actually used it.
- Actual Copying
Actual copying means that a defendant did in fact use the copyrighted work in creating the impugned work. This can be proved directly, through evidence that the defendant copied the work, or indirectly, by showing the defendant had access to it and produced something similar.
In its decision, the Court compared how similar 2,830 Bulk Memos, Headnotes and judicial opinions were one by one, acknowledging that LegalEase had access to Westlaw and used it to create the Bulk Memos, while noting that access alone is not proof of copying. The Court further observed that in instances where a Bulk Memo question looked more like a Headnote than an underlying judicial opinion there was strong circumstantial evidence of actual copying.
Upon the Court’s review of the 2,830 Headnotes, they granted summary judgment to Thomson Reuters on actual copying on 2,243 of them – a subset of which actual copying was so obvious that no reasonable jury could find otherwise.
- Substantial Similarity
The substantial similarity analysis requires an evaluation of whether the latter work materially appropriates the copyrighted work, and involves a consideration of which parts of the actually copied work are original expressions protected by copyright. In considering this, the Court noted that the question is whether an ordinary user of a product would find it substantially similar to the copyrighted works.
Ross argued that the Bulk Memo questions had to not just be substantially similar to the Headnotes, but virtually identical due to the “thin” nature of the Headnotes’ underlying copyrights. Effectively, the less protectable expression a work contains, the more similar the impugned work must be to it.
Applying this standard, the Court granted summary judgment on substantial similarity on 2,243 Headnotes, finding that the Bulk Memo questions were substantially similar to them. In this regard, the Court specified that it granted summary judgment only on the Headnotes whose language “very closely track[ed] the language of the Bulk Memo question but not the language of the case opinion”.
- Ross Defenses to Copyright Infringement Failed
The Court rejected all of Ross’ defenses of innocent infringement, copyright misuse, merger and scenes a faire.
The Court summarily dismissed Ross’ claim that any infringement was innocent. The defence of innocent infringement acts to only limit damages, as opposed to liability. However, Thomson Reuters’ Headnotes included a copyright notice, and as such innocent infringement did not apply.
The Court also dismissed Ross’ argument that Thomson Reuter misused its copyright, alleging it weaponized its copyright against the public interest, a form of “anti-competitive behaviour”. The Court disagreed with this contention, and found that Ross had not shown that Thomson Reuters misused its copyrights to “stifle competition”.
Ross further claimed the defense of merger, alleging that any ideas embodied in the Headnotes were so close to the expressions that they merged with the expressions themselves, in turn making the Headnotes uncopyrightable. The Court rejected this defense, having found that copyright subsisted in the Headnotes and holding that there were a variety of ways to express points of law from judicial opinions.
The Court also rejected Ross scenes a faire defence, which concerns “stock elements that follow from the work’s nature” – essentially, common tropes associated with a work. In this regard the Court observed that “nothing about a judicial opinion requires it to be slimmed down” in the same manner as Thomson Reuters’ Headnotes.
- Thomson Reuters Prevailed on Fair Use
In assessing Ross’ claims of fair use, the Court considered the four factors required to establish the defence of fair use, including:
- the use’s purpose and character, including whether it was commercial or nonprofit;
- the nature of the copyrighted work;
- how much of the work was used and how substantial a part it was relative to the copyrighted work’s whole; and
- how Ross’ use affected the copyrighted work’s value or potential market.
Although fair use is a question of mixed fact and law, the Court held that the undisputed facts of the case pushed it “squarely into the legal realm” such that the fair use question was appropriate for a judge to determine and not a jury.
The Court ultimately held that Thomson Reuters prevailed on factors 1 and 4, the two most important factors, and on the factors’ overall balancing.
Factor 1 – The Use’s Purpose and Character was Commercial and Non-Transformative
With respect to the first factor, Court found that Ross’ use was commercial in nature and not transformative. The Court recognized Ross’ admission that it stood to profit from the exploitation of the copyright protected material without paying the customary price, and was commercial in nature.
The Court further found that Ross’ use was not transformative as it did not have a “further purpose or different character” from the Thomson Reuters’ Headnotes. However, Ross argued that because the Headnotes did not appear as part of Ross’ final product, the copying therefore occurred at intermediate steps, and that intermediate copying had been permitted under the first fair use factor in prior decisions concerning computer programs.
The Court rejected this contention, distinguishing the case from previous intermediate-copying decisions on the basis that computer programs differ from other literary works as computer programs always serve a functional purpose, and that intermediate-copying in such cases was necessary for the competitors to innovate. In the present case, the copying was not found to be reasonably necessary to achieve Ross’s new purpose.
Bad Faith Irrelevant and Immaterial
Due to Ross’ commercial and non-transformative use, the Court did not consider whether Ross had acted in bad faith.
Factor 2 – The Headnotes Had Minimal Creativity
The Court’s analysis on the nature of the original work focused on the degree of creativity inherent to the Headnotes. It found that Thomson Reuters’ material had more than the minimal spark of originality required for copyright to subsist, but that the Headnotes were not especially creative. As such, the Court found in Ross’ favour on this factor, but qualified this finding noting that the second factor “rarely play[s] a significant role in the determination of a fair use dispute”.
Factor 3 – Ross’ Use Did Not Constitute a Substantial Part Relative to the Whole of the Headnotes
Under factor 3 of the analysis, the Court assessed whether Ross’ use was “reasonable in relation to the purpose of the copying”, considering the “quantity of the materials used’ and “their quality and importance”. The question core to this analysis was whether Ross had copied the “heart” of the works at issue.
In undertaking the analysis, the Court noted that what matters is the amount and substantiality of what was made accessible to a public which may serve as a substitute. As Ross did not make the Headnotes available to the public, the Court found in Ross’ favour.
Factor 4 – Ross’ Use Affected the Headnotes’ Value or Potential Market
Under the fourth factor, the Court found in favour of Thomson Reuters, holding that Ross meant to compete with Thomson Reuters by creating a market substitute, and that Ross had not put forward sufficient facts to show that a potential market for AI training data did not exist and would not be affected by Ross’ use.
With respect to Ross’ arguments concerning the public interest, the Court further found that public interest in the subject matter at issue alone was insufficient, and that the public had no right to Thomson Reuters’ “parsing of law”.
The Court’s Findings
The Court ultimately held that Ross had infringed 2,243 Headnotes (and the only remaining factual issue was whether some of their copyrights had expired) and that Ross’ innocent infringement, copyright misuse, merger and scenes a faire defenses all failed.
Key Takeaways
While uncertainty remains as to how Canadian courts would decide a similar case, this decision may be an indication of the perils of training AI on copyright protected works. In any event, the decision reinforces the protections conferred by copyright ownership and serves as a reminder of the importance of entering into valid licensing agreements before training AI on copyright protected works.