Robert Ambrogi Blog
As the long-running litigation between Thomson Reuters and ROSS Intelligence has moved to the 3rd U.S. Circuit Court of Appeals for an interlocutory appeal of the trial judge’s copyright rulings in favor of TR, amicus curiae have come out in force to support the position of the now-defunct AI legal research startup that it did not violate copyright law.
Ten amicus briefs have been filed, all arguing in favor of ROSS. The various amici represented in the briefs include upcoming legal research startups, civil liberties and open-information groups, industry trade associations, computer scientists, law professors, and others.
Many of the briefs focus on the proposition, “No one can own the law,” as the Supreme Court said in the 2020 case Georgia v. Public.Resource.Org. That is true of judicial opinions and should also be true of headnotes, say several of the briefs.
As lawyer Phillip R. Malone and law student August Gebhard-Koenigstein of Stanford Law School put it in their brief on behalf of legal research and technology startups, “The same should be true of the headnotes at issue in this case, which serve a discrete and limited purpose as often near-verbatim summaries and verbatim quotes that faithfully and accurately describe a specific point of law from a judicial opinion.”
Last February, the trial judge in the case, U.S. Circuit Judge Stephanos Bibas, sitting by designation in the U.S. District Court in Delaware, issued partial summary judgment in favor of Thomson Reuters on two key issues of copyright law. He then postponed the scheduled trial to allow ROSS to file an interlocutory appeal on those two issues.
The two issues Judge Bibas certified for appeal to the 3rd Circuit are:
- Are Westlaw’s editor-created headnotes and Key Number system sufficiently original to qualify for copyright protection? The answer could have profound implications for the legal publishing industry. Judge Bibas concluded that attorney editors exercise enough creativity when they choose which parts of judicial opinions to highlight and how to categorize them, meeting copyright law’s low bar for originality.
- Fair use. Even if the headnotes are copyrightable, did ROSS’s use of them constitute fair use under copyright law? Judge Bibas found that ROSS’s use was not “transformative” because both companies created legal research tools that “identify the important parts of a large body of law.” He concluded that ROSS essentially created a “market substitute” for Westlaw rather than adding new meaning or purpose to the original work.
Since then, these 10 amicus briefs have been filed. All favor ROSS. No amicus briefs have been filed in favor of TR.
Below are quick summaries of the amicus briefs, including a “nugget” quote, with links to the full text of each.
Amicus Briefs Summary: Thomson Reuters v. ROSS Intelligence
1. Heather Meeker
Identity and Background: Heather Meeker, partner at Technology Law Partners LLP, is a preeminent software licensing and copyright attorney known for her work on open-source and AI legal issues.
Counsel for Amicus: Heather Meeker, Scotts Valley, Calif.
Supports: ROSS Intelligence
Principal Arguments:
- Headnotes lack originality. They simply restate court holdings and therefore are factual, not creative.
- Merger doctrine. Any expression merges with underlying legal ideas.
- Public domain. Extending copyright to case law summaries privatizes public information.
- Collective-work issue. Even if copyrightable, headnotes are only protectable as part of the collective database, not as 2,243 separate works.
Nugget Quote: “Headnotes do not enjoy copyright protection. They fail to meet the fundamental requirements under federal law, because they lack sufficient originality, creativity, and expression to qualify for copyright protection. They are formulated intentionally to be factual in nature, and to track the language of their source, and they are based on materials in the public domain – mainly court opinions – whose availability to the body politic is a fundamental precept of liberty in a nation of laws.”
2. Electronic Frontier Foundation, ALA, ARL, Internet Archive, Public Knowledge, and Public.Resource.Org
Identity and Background: Six leading civil-liberties and open-information groups long active in defending public access to law and digital knowledge.
Counsel for Amicus: Victoria J. Noble, Electronic Frontier Foundation, San Francisco, Calif.
Supports: ROSS Intelligence
Principal Arguments:
- ‘No one can own the law.’ Judicial opinions and their summaries are public domain under the government edicts doctrine.
- Minimal originality. Headnotes are largely factual or functional and therefore unprotectable.
- Fair use. Even if copyrightable, AI training is a transformative, factual use that serves the public interest.
- Public benefit. Restricting use would expand private control over legal materials.
Nugget Quote: “Copyright law does not prohibit the use of judicial decisions to build useful legal research tools. It does not empower private companies to monopolize access to the law. Yet the district court’s decision granting summary judgment to Thomson Reuters on infringement of its Westlaw headnotes would expand private control over the law and increase the public’s costs of access.”
3. Computer & Communications Industry Association (CCIA), Chamber of Progress, and NetChoice
Identity and Background: Trade associations representing major technology and Internet firms, advocating balanced copyright policy and open markets.
Counsel for Amicus: Michael S. Kwun and Elizabeth H. Dinh, Kwun Bhansali Lazarus LLP, San Francisco, Calif.
Supports: ROSS Intelligence
Principal Arguments:
- Not generative AI. ROSS used AI for legal search, not content creation; courts should not overgeneralize AI cases.
- Transformative fair use. ROSS’s copying was an intermediate step to train an AI search tool, not to substitute for Westlaw.
- Market effect. Functional competition is not market substitution; the products differ fundamentally.
- Precedents. Analogous to Google v. Oracle, HathiTrust, and Perfect 10.
- Policy caution. Courts should avoid rulings that chill legitimate AI innovation.
Nugget Quote: “While this is one of the first appellate proceedings regarding fair use and artificial intelligence (AI), the AI at issue in this appeal is very different from the generative AI models that are being considered in other litigation. Amici request that this Court take care that its decision does not sweep unintentionally broadly. Issues specific to generative AI models are not before this Court and should be decided on their own merits.”
4. Randy Goebel and Larry Ullman
Identity and Background: Randy Goebel, Professor of Computing Science, University of Alberta, an AI researcher specializing in legal reasoning; Larry Ullman, noted programming author and AI documentation consultant.
Counsel for Amicus: Michael P. Abate and Burt A. (Chuck) Stinson, Kaplan Johnson Abate & Bird LLP, Louisville, Ky.
Supports: ROSS Intelligence
Principal Arguments:
- Court misunderstood AI. The lower court’s “generative vs. non-generative” dichotomy is scientifically wrong.
- ROSS’s AI is transformative. It builds a semantic “brain” to analyze case law — functionally generative in purpose.
- Purpose shift. ROSS uses text to teach machines meaning, not to replicate expression.
- Policy warning. Oversimplified AI distinctions could hinder legal AI development.
Nugget Quote: “Generative AI examines vast amounts of data, applies pattern recognition and other human-like learning techniques, and builds a virtual encyclopaedic ‘brain’ of expertise in a given field. This brain can then be used to create new content within that field: an image, a poem, a recipe. But — here’s the key — the generative aspect includes creating the brain, unbeknownst to the eventual end user. As explained below, that’s precisely what ROSS’s AI does.”
5. Next-Generation Legal Research and Technology Platforms (Cicerai, Trellis Research, Paxton AI, JuristAI, Free Law Project, and Dispute Resolution AI)
Identity and Background: A coalition of emerging legal-tech innovators represented by Stanford’s Juelfsgaard Intellectual Property and Innovation Clinic.
Counsel for Amicus: Phillip R. Malone and August Gebhard-Koenigstein (Certified Law Student), Stanford Law School, Calif.
Supports: ROSS Intelligence
Principal Arguments:
- Headnotes lack originality. They are factual restatements, not creative expression.
- Merger doctrine. Legal ideas and their phrasing cannot be separated.
- Fair use. Model training is transformative and comparable to Google Books and HathiTrust.
- Public interest. Limiting AI use would harm access, innovation, and competition in legal information markets.
Nugget Quote: “Allowing overbroad assertions of copyright over headnotes, or denying fair use protection for using such materials for intermediate, transformative purposes like training new AI models or developing groundbreaking AI systems, undermines the public interest and will exacerbate existing barriers to legal information, impede innovation, and reduce competition, perpetuating the dominance of the current few major players.”
6. Authors Alliance
Identity and Background: Nonprofit of 3,000+ academic and creative authors advocating for fair use and dissemination of scholarship.
Counsel for Amicus: Karen E. Keller, Shaw Keller LLP, Wilmington, Del.
Supports: ROSS Intelligence
Principal Arguments:
- Intermediate copying doctrine. Fair use extends beyond computer code to AI training and text/data mining.
- Distinct purpose. ROSS’s use (machine learning) differs fundamentally from Westlaw’s (human reading).
- Market focus. Harm must be to an expressive licensing market, not speculative AI training markets.
- Public value. Supports access to legal information and research freedom.
- Bad faith rejected. Refusal of a license cannot defeat fair use.
Nugget Quote: “ROSS’s use serves compelling public interests by increasing access to legal information through innovative research tools, while its outputs consist entirely of public domain judicial opinions that cannot substitute for Westlaw’s proprietary content. … This Court should preserve the vitality of fair use for all creators, researchers, and innovators and reverse the district court’s contrary ruling.”
7. Professors Brian L. Frye, Jess Miers, and Mateusz Blaszczyk
Identity and Background: Law professors and researchers focused on copyright, technology, and AI policy.
Counsel for Amicus: Michael T. Hensley and Jeffrey Miles, Carlton Fields, P.A., Florham Park, N.J., and Los Angeles, Calif.
Supports: ROSS Intelligence
Principal Arguments:
- Headnotes are facts. They merely describe judicial holdings and are uncopyrightable.
- Key Number System public domain. Created in 1909, its copyright expired decades ago.
- Fair use. ROSS’s AI training parallels Sega and Connectix reverse-engineering cases.
- Reject “market dilution.” Copyright does not shield against functional or stylistic competition.
- Public ownership of law. Privatizing legal ideas violates Banks v. Manchester and due process.
Nugget Quote: “What ROSS’s legal experts did to prepare the training memos was no different from what lawyers do when preparing briefs like this one. Yet it is this ordinary legal research task — and the expert-authored work it produced — that Thomson Reuters would paint as infringement with a broad brush, without pausing to make the threshold determination as to whether specific holdings expressed in each headnote that flowed into this process were themselves protectable.”
8. Copyright Law Professors (Tushnet, Samuelson, Sag, Sprigman, and Lee)
Identity and Background: Five leading U.S. copyright scholars from Harvard, Berkeley, NYU, Emory, and Santa Clara.
Counsel for Amicus: Rebecca Tushnet, Harvard Law School, Cambridge, Mass.
Supports: ROSS Intelligence
Principal Arguments:
- Transformative purpose. ROSS used headnotes to train AI, not to convey expressive content; similar to Google Books and HathiTrust.
- Warhol misapplied below. The district court wrongly equated business competition with expressive substitution.
- Thin copyright. Headnotes are factual and constrained; at most minimally creative.
- Amount used. Limited and reasonable for functional training.
- Market harm. No substitution or expressive usurpation—ROSS outputs only judicial opinions.
- Public benefit. Expands access to law and competition in legal research.
Nugget Quote: “[P]eople benefit from new search tools that enhance their ability to find information. Indeed, ROSS’s tool provides a public benefit of the highest order in our democracy by fostering an informed citizenry with greater accessibility to judicial opinions. ROSS’s nonexpressive, highly transformative use is a fair use.”
9. Foundation for American Innovation
Identity and Background: A nonprofit think tank promoting innovation and technology policy, founded by technologist and policy scholar Tim Hwang.
Counsel for Amicus: Tim Hwang, San Francisco, Calif.
Supports: ROSS Intelligence
Principal Arguments:
- Intermediate copying justified. Nonpublic, functional training satisfies Warhol’s “other justification” for fair use, like Sega, Connectix, and Google v. Oracle.
- Factor one. Copying was limited to what was reasonably necessary to train the AI; outputs excluded any headnote text.
-
Factor four. Only expressive licensing markets count; “training licenses” and platform competition are non-cognizable.
-
Policy. Treating AI training as infringement would chill innovation and entrench incumbents.
- Rule proposed. Nonpublic, necessary functional training = fair use; only expressive substitution is actionable.
Nugget Quote: “Treating nonpublic ‘training’ as a cognizable market would misalign copyright with its constitutional purpose of promoting progress by protecting expression while leaving ideas, facts, and methods free for all. It would balkanize datasets, raise fixed compliance costs, and entrench incumbents at the expense of research labs, nonprofits, and startups.”
10. Abraham Kang, Esq., and Kunal Patel
Identity and Background: Kang is an attorney and AI technologist; Patel a machine learning engineer. They filed independently to explain AI’s technical realities.
Counsel for Amicus: Abraham Kang, Los Gatos, Calif.
Supports: ROSS Intelligence
Principal Arguments:
- District court analyzed wrong act. The fair use inquiry should focus on the transformative creation of vector embeddings, not downstream product competition.
- AI training = functional transformation. Like Sega and Connectix, copying to extract unprotected functionality is fair use.
- Warhol alignment. ROSS’s use had a distinct purpose (machine learning) versus Westlaw’s (human reading); no substitution.
- Market harm speculative. “AI training data” market is not a traditional or reasonable derivative market.
- Public benefit. ROSS broadened access to law; ruling against it would weaponize copyright against technological progress.
Nugget Quote: “The district court’s decision rests on a cascade of errors rooted in a misunderstanding of artificial intelligence. Its fundamental error, which infects its entire fair use analysis, was analyzing the wrong act. By focusing on the downstream, competitive outputs of ROSS’s search tool instead of the initial, direct transformation of copyrighted text into mathematical vector embeddings, the court applied the fair use factors to the wrong conduct entirely.”