694 F.Supp.3d 467
D. Del.2023Background
- Thomson Reuters owns Westlaw, which organizes judicial opinions with a Key Number System and author-written headnotes; Westlaw has a registered copyright covering its compilation and headnotes.
- Ross Intelligence built an AI legal-research engine and hired LegalEase to produce a "Bulk Memo Project" (~25,000 Q&A sets) consisting of lawyer-style questions and quoted opinion passages; Ross used those memos to train ML models (encoding, featurizer, etc.).
- Dispute centers on whether LegalEase copied Westlaw headnotes (Thomson Reuters asserts 2,830 clearly-copied instances for summary judgment) and whether Ross is liable for using those memos to train its AI.
- Ross also received (and considered) a list of 91 Westlaw Key Number topics and 500 opinions (with headnotes); Ross says it did not use the 500 opinions in training and abandoned a classifier project based on the topics.
- Procedural posture: cross-motions for summary judgment on copyright infringement (limited to the 2,830 memos), fair use, tortious interference with contract, and Ross’s preemption defense; many central facts are genuinely disputed and sent to the jury.
Issues
| Issue | Plaintiff's Argument (Thomson Reuters) | Defendant's Argument (Ross) | Held |
|---|---|---|---|
| Validity/scope of Westlaw copyright (compilation vs. individual headnotes & Key Number System) | Westlaw’s registration covers its compilation and the individually copyrightable headnotes and Key Number System. | The registration is a compilation; copying a few thousand entries from hundreds of thousands cannot support infringement; Key Number System is unoriginal or minimally creative. | Validity and scope are fact questions for the jury: Key Number System originality and headnotes’ independence remain disputed. |
| Actual copying / substantial similarity (the 2,830 memos) | LegalEase admitted copying portions; access is undisputed; substantial similarity exists for many entries. | Ross contends questions were independently authored or mirror underlying opinions (not headnotes); large overlap may derive from public-domain or uncopyrightable opinion text. | Actual copying (use of Westlaw material) is established as a matter of law for portions; substantial similarity largely is a jury question (some entries undisputedly similar but protection remains factual). |
| Liability theories (direct, contributory, vicarious) | Ross directly reproduced and hosted memos; Ross knew and materially assisted LegalEase; Ross had practical control and financial interest. | Ross denies knowledge of infringement, denies instruction to copy, and disputes ability to supervise LegalEase. | All theories raise factual disputes (knowledge, control, material contribution) and must go to trial; cannot be resolved on summary judgment. |
| Fair use defense | Ross’s use is commercial and competes with Westlaw; translation into numeric form and training were not transformative but derivative. | Training an AI by intermediate copying is transformative; Ross’s final outputs do not contain the copied expressive material; use resembles Sega/Connectix intermediate-copying fair uses. | Fair-use factors present disputed subsidiary facts (transformativeness, amount, market effect); fair use must be decided by a jury. |
| Tortious interference / preemption (contract provisions) | Ross induced LegalEase to breach contractual provisions barring bulk use, bots, and password sharing. | Federal copyright preempts state claims that are equivalent to §106 rights; some contract provisions merely limit use and are not preempted. | Anti-competition (bulk/replace-substitute) claim is preempted by Copyright Act; anti-bot and password-sharing interference claims are not preempted and survive to trial (partial SJ granted on existence of contract and harm). |
Key Cases Cited
- Feist Publ’ns, Inc. v. Rural Tel. Serv. Co., 499 U.S. 340 (1991) (idea–expression distinction and copyright elements)
- Anderson v. Liberty Lobby, Inc., 477 U.S. 242 (1986) (summary judgment standard—genuine dispute for jury)
- Google LLC v. Oracle Am., Inc., 141 S. Ct. 1183 (2021) (fair use is mixed law/fact; technological-context guidance)
- Authors Guild v. Google, Inc., 804 F.3d 202 (2d Cir. 2015) (intermediate copying and public-benefit considerations in fair use)
- Sega Enters. Ltd. v. Accolade, Inc., 977 F.2d 1510 (9th Cir. 1992) (intermediate copying for functional/transformative purpose)
- Sony Computer Ent. Inc. v. Connectix Corp., 203 F.3d 596 (9th Cir. 2000) (reverse engineering/intermediate copying fair use)
- Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994) (transformativeness and commerciality in fair use)
- Harper & Row Publishers, Inc. v. Nation Enters., 471 U.S. 539 (1985) (market harm and core copyright protection)
- Petrella v. Metro-Goldwyn-Mayer, Inc., 572 U.S. 663 (2014) (laches not available to bar copyright claim)
- Tanksley v. Daniels, 902 F.3d 165 (3d Cir. 2018) (actual copying standard and substantial-similarity framework)
- Leonard v. Stemtech Int’l Inc., 834 F.3d 376 (3d Cir. 2016) (standards for contributory and vicarious liability)
- Dun & Bradstreet Software Servs., Inc. v. Grace Consulting, 307 F.3d 197 (3d Cir. 2002) (extra-element test for copyright preemption)
- MAI Sys. Corp. v. Peak Comput. Inc., 991 F.2d 511 (9th Cir. 1993) (hosting copies can constitute reproduction)
- WaveDivision Holdings, LLC v. Highland Cap. Mgmt., L.P., 49 A.3d 1168 (Del. 2012) (elements of tortious interference with contract)