The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Thomson Reuters won a major trial-level victory in the United States’ first prominent AI-training fair-use dispute on February 11, 2025. A Delaware federal judge ruled that Ross Intelligence infringed copyrights in 2,243 Westlaw headnotes and rejected Ross’s fair-use defense.
But the decision is narrower than “AI training is illegal.” It involved a non-generative legal-research system, specifically copied Westlaw editorial content, and a product designed to compete with Westlaw. The ruling is also under interlocutory appeal in the U.S. Court of Appeals for the Third Circuit, with no appellate decision identified as of August 18, 2026.
| # | Preview | Product | Price | |
|---|---|---|---|---|
| 1 |
|
Copyright Law: Cases and Materials (v8.0) | $19.87 | Buy on Amazon |
| 2 |
|
Music Copyright Law | $32.95 | Buy on Amazon |
| 3 |
|
Understanding Copyright Law (Understanding Series) | $55.00 | Buy on Amazon |
| 4 |
|
Copyright Law | $33.62 | Buy on Amazon |
| 5 |
|
Copyright Law in a Nutshell | $65.00 | Buy on Amazon |
As an Amazon Associate I earn from qualifying purchases.
What Thomson Reuters actually won
In Thomson Reuters Enterprise Centre GmbH and West Publishing Corp. v. Ross Intelligence Inc., Judge Stephanos Bibas of the U.S. District Court for the District of Delaware granted most of Thomson Reuters’ motion for partial summary judgment on important liability issues.
The court found that Ross had copied protected material from Westlaw while developing an AI-powered legal-research product intended to compete with Westlaw. Of 2,830 examined headnotes, the judge found actual copying involving 2,243. He also rejected Ross’s fair-use defense.
#1 Best Overall
The ruling was significant because it addressed whether copyrighted material used in developing an AI system could qualify as fair use. It was widely characterized as the first major U.S. ruling squarely addressing fair use in AI-training litigation. It was not, however, a Supreme Court decision, a nationwide rule, or a ruling that all commercial AI training infringes copyright.
Read the February 11, 2025 district-court opinion.
How the dispute began
Thomson Reuters and West Publishing sued Ross Intelligence in 2020. Westlaw is Thomson Reuters’ legal-research platform; Ross was developing a competing AI-assisted legal-research service.
The dispute was not primarily about copying judicial opinions. Court opinions themselves are generally public-domain government works. Instead, it concerned Westlaw’s editorial content—especially its headnotes and the related Key Number System.
A Westlaw headnote is a concise editorial statement identifying a legal proposition from a longer judicial opinion. West’s editors choose which propositions to highlight, characterize them, phrase them, and organize them within Westlaw’s research system.
According to the district-court record, Ross used an outside vendor called LegalEase to create “Bulk Memos”: legal questions and answers allegedly derived from Westlaw headnotes. Ross then used those materials in developing its legal-search system. Ross later ceased operations, reportedly in 2021, but the litigation continued.
Why the court found the headnotes copyrightable
Ross argued, in substance, that the headnotes described law and therefore consisted of uncopyrightable facts or public legal principles. The court distinguished between the underlying law and West’s editorial expression of it.
Rank #2
The legal rule contained in a headnote is not owned by Thomson Reuters merely because Westlaw summarizes it. But a sufficiently original selection, arrangement, and expression of that rule can receive copyright protection. The court concluded that the relevant headnotes contained enough editorial choice to qualify for protection, even though their subject matter was factual or legal.
PC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteThat protection is relatively narrow. It does not give Thomson Reuters ownership of court decisions or exclusive rights over every way of explaining a legal principle. It does mean that copying the protected wording or editorial choices can create infringement risk.
Why Ross’s fair-use defense failed
U.S. fair use is evaluated under four factors. Judge Bibas considered the factors together, with the specific facts of Ross’s development process and intended product shaping the result.
1. Commercial and competitive purpose
Ross was developing a commercial legal-research service that would compete directly with Westlaw. The court viewed that purpose as weighing against fair use.
The important qualification is that the court did not hold that commercial AI training is automatically unlawful. The competitive relationship mattered because Ross used Westlaw’s editorial material to help create a product serving the same legal-research market.
2. Nature of the copyrighted works
The headnotes concerned legal propositions, making them more factual and constrained than a novel, photograph, or other highly creative work. That factor could support fair use to some extent.
But “thin” copyright protection is not the same as no protection. The court found that Westlaw’s editorial selection and wording contained protectable expression. A defendant may infringe a thinly protected work by copying the expression that copyright does protect.
3. Amount and substantiality of the copying
The court examined thousands of headnotes and found actual copying involving 2,243 of the 2,830 headnotes reviewed. That was not treated as a small number of incidental references.
The figure is a case-specific finding, not a legal threshold. Copyright law does not say that copying more than a particular number of items is automatically infringement. The significance of the amount depends on what was copied, how much of it was copied, and its importance to the defendant’s use.
4. Effect on the market
The direct competition between Ross’s intended product and Westlaw was central. The court treated Ross’s system as a potential substitute for the service whose editorial content had been copied.
This market-substitution analysis may not transfer neatly to every general-purpose AI model. A broad model trained on many sources, producing open-ended outputs, may have a different relationship to any particular source than Ross’s legal-search system had to Westlaw. Whether a source’s market is harmed, and how, will depend on the facts of each dispute.
What the ruling did not decide
| This case involved | It did not decide |
|---|---|
| A non-generative AI legal-search system | Whether all generative AI training is lawful or unlawful |
| Specific Westlaw headnotes and editorial content | Whether every copyrighted training corpus may be used without permission |
| A product intended to compete directly with Westlaw | Whether every commercial AI product is a substitute for its training sources |
| Documented copying of particular materials | Whether every intermediate training copy infringes |
| A Delaware federal district-court ruling | A nationwide Supreme Court rule |
Why this was not a generative-AI case
Ross’s product was an AI-powered legal-research tool, not a chatbot or image generator producing open-ended content. The district court therefore did not decide the legality of training systems such as large language models, image-generation systems, or other general-purpose generative products.
Rank #4
The distinction matters because the product’s function, source material, development record, and relationship to the source market all influence fair-use analysis. A legal-search system trained with copied headnotes to compete with Westlaw presents a different factual question from a general model trained on a large and varied corpus.
Recommended Free Tools
The ruling may influence lawsuits involving generative AI, particularly arguments about intermediate copying, market substitution, and the use of expressive copyrighted material. But influence is not the same as a controlling answer.
Intermediate copying is not automatically protected
Ross’s position included the argument that copying during development was an intermediate step and that the final product did not simply reproduce Westlaw’s headnotes. The court rejected that defense on the record before it.
Intermediate copying can sometimes be lawful, depending on the purpose, necessity, transformation, applicable precedent, and effect on markets. It is not automatically permissible merely because the copied material is not displayed verbatim in the final product.
For AI developers, the practical lesson is that the complete development chain may matter: where training or reference material came from, whether it was copied, how it was used, what product it helped create, and whether that product competes with the source.
Practical implications for AI companies and rights holders
Training-data provenance matters
AI developers face greater legal risk when they cannot explain the origin and permitted use of training or development materials. Provenance records, licensing agreements, access controls, and documented independent-development processes can become important evidence in litigation.
Best Value
Public-domain source material is not the whole analysis
Developers may generally work with public-domain judicial opinions, but a commercial database can add protectable editorial layers around those opinions. The distinction between public source material and copyrighted summaries, annotations, organization, or other editorial enhancements deserves careful attention.
Direct competition can intensify fair-use risk
The ruling suggests that copying expressive material to build a product that substitutes for the source service can be especially problematic. That is an implication from this decision, not a categorical rule for every AI product.
Licensing may reduce uncertainty
Rights holders may respond by licensing data, imposing contractual restrictions, or offering authorized AI products. Developers may decide that licensing costs are preferable to litigation uncertainty, although the commercial and legal value of any license depends on the data, jurisdiction, and intended use.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Users still need primary-source verification
Whether an AI tool is trained on licensed material does not guarantee that every answer is correct. Legal users should still verify outputs against the underlying opinion, statute, regulation, filing, or docket entry.
The appeal and what happens next
The district-court ruling is not the end of the case. The relevant procedural timeline is:
- February 11, 2025: Judge Bibas issued the district-court ruling.
- June 17, 2025: The Third Circuit granted permission for an interlocutory appeal.
- June 24, 2025: The appeal was transferred to merits docket No. 25-2153.
- June 11, 2026: The Third Circuit heard oral argument.
- As of August 18, 2026: No appellate decision had been identified in the cited docket and coverage.
The Third Circuit could affirm the ruling, reverse or narrow the copyrightability analysis, reverse or narrow the fair-use decision, or send issues back to the district court. It may address the certified legal questions without resolving every remaining issue, including any eventual remedy.
The available sources do not establish a final damages award. Thomson Reuters’ “win” should therefore be described as a victory on key district-court infringement and fair-use issues, not as a verified monetary recovery or an unreviewable final judgment.
Do these 3 things before closing this tab:
1Scan for outdated or missing drivers - takes under a minute2Clear out junk files and repair common Windows errors3Fix the driver behind crashes, sound loss and screen glitchesThe bottom line
Thomson Reuters achieved a major rights-holder victory: a federal district judge found that Ross copied protected Westlaw headnotes to help build a competing legal-research AI system and held that the use was not fair use on those facts.
That is an important warning for AI developers, especially those using copyrighted editorial content in products that directly compete with the source. It is not a universal ruling that AI companies cannot train on copyrighted material, nor does it decide the legality of all generative-AI models. The Third Circuit’s pending appeal is the next major development.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
