# Thomson Reuters v. Ross Intelligence: Delaware court rejects fair use for AI training data, finds 2,243 Westlaw headnotes infringed

> On February 11, 2025, U.S. District Judge Stephanos Bibas (sitting by designation, D. Del.) granted partial summary judgment to Thomson Reuters, holding that legal-AI startup Ross Intelligence infringed copyright in 2,243 Westlaw headnotes and that its copying to train a legal-research AI was not fair use — the first major U.S. ruling to reject a fair-use defense for AI training data.

- Verification status: verified
- Case type: deployment
- Provider: Ross Intelligence Inc. (legal-research AI trained on Westlaw headnotes via LegalEase 'Bulk Memos')
- Client: U.S. District Court for the District of Delaware (Bibas, Circuit Judge, sitting by designation) — Thomson Reuters Enterprise Centre GmbH & West Publishing Corp. v. Ross Intelligence Inc., No. 1:20-cv-613-SB, Courts / legal (AI-copyright / fair-use adjudication) (named)
- Sector: legal-services / US / cross
- Verified on: 2026-08-19
- Canonical URL: https://theinternetninja.com/stories/thomson-reuters-v-ross-intelligence-dde-bibas-2243-westlaw-headnotes-no-fair-use-summary-judgment-2025/
- Source: The Internet Ninja (theinternetninja.com), independent verified-proof platform

## Outcomes

| Metric | Before | After |
| --- | --- | --- |
| Partial summary judgment for Thomson Reuters: Ross found to have directly infringed copyright in 2,243 Westlaw headnotes |  |  |
| Fair-use defense rejected — factors one (non-transformative, commercial) and four (market harm) favor Thomson Reuters |  |  |
| First major U.S. court ruling to reject a fair-use defense for using copyrighted works as AI training data |  |  |

## Verification method

Signed federal Memorandum Opinion (D. Del., Feb 11 2025, D.I. 770), byte-tied to its GovInfo Wayback captures (CDX digest match); quotes verbatim from the opinion; corroborated firsthand by two independent newsrooms, WIRED and The Register (both byte-tied to organic Wayback captures).

## Full case file

## The problem
AI companies routinely train models on copyrighted material and defend the practice as
"fair use" — arguing the copying is transformative and does not substitute for the
original. Until 2025 no U.S. court had squarely decided whether that defense holds when
the copied works are used as AI training data. *Thomson Reuters Enterprise Centre GmbH v.
Ross Intelligence Inc.* put the question to a federal judge: Ross, a legal-research AI
startup, had trained its search tool on "Bulk Memos" — question-and-answer sets a
contractor (LegalEase) built from Westlaw's editorial headnotes — and Thomson Reuters,
Westlaw's owner, sued for copyright infringement in 2020 [source](https://www.wired.com/story/thomson-reuters-ai-copyright-lawsuit/).

## What was built
Ross Intelligence built a legal-research search tool powered by AI. As the court
recounted, LegalEase gave lawyers "a guide explaining how to create those questions using
Westlaw headnotes, while clarifying that the lawyers should not just copy and paste
headnotes directly into the questions," and the resulting Bulk Memos were used to train
Ross's system [source](https://www.govinfo.gov/content/pkg/USCOURTS-ded-1_20-cv-00613/pdf/USCOURTS-ded-1_20-cv-00613-5.pdf). The court was careful to note the tool's nature: "It is undisputed
that Ross's AI is not generative AI" — rather, "when a user enters a legal question, Ross
spits back relevant judicial opinions that have already been written" [source](https://www.govinfo.gov/content/pkg/USCOURTS-ded-1_20-cv-00613/pdf/USCOURTS-ded-1_20-cv-00613-5.pdf). Revising his
own 2023 ruling, the judge wrote: "I thus revise my 2023 summary judgment opinion and
order in this case" [source](https://www.govinfo.gov/content/pkg/USCOURTS-ded-1_20-cv-00613/pdf/USCOURTS-ded-1_20-cv-00613-5.pdf).

## The outcome
**The number that gets quoted: <span class="kpi">2,243</span> Westlaw headnotes found infringed.**
**The precedent: the <span class="kpi">first</span> major U.S. ruling to reject fair use for AI training data (press characterization, not a self-award in the opinion).**

In a Memorandum Opinion filed February 11, 2025, Judge Stephanos Bibas — a Third Circuit
judge sitting by designation in the District of Delaware — held: "Ross infringed <span class="kpi">2,243</span>
headnotes" and "None of Ross's possible defenses holds water. I reject them all" [source](https://www.govinfo.gov/content/pkg/USCOURTS-ded-1_20-cv-00613/pdf/USCOURTS-ded-1_20-cv-00613-5.pdf). On
fair use, he found Ross's use commercial and "not transformative because it does not have
a 'further purpose or different character' from Thomson Reuters's," and held that the
market-effect factor — "undoubtedly the single most important element of fair use" —
favored Thomson Reuters, expressly including a derivative market for "data to train legal
AIs": "The original market is obvious: legal-research platforms. And at least one
potential derivative market is also obvious: data to train legal AIs" [source](https://www.govinfo.gov/content/pkg/USCOURTS-ded-1_20-cv-00613/pdf/USCOURTS-ded-1_20-cv-00613-5.pdf). Tallying
the factors — "Factors one and four favor Thomson Reuters. Factors two and three favor
Ross" — he concluded, "Weighing them all together, I grant summary judgment for Thomson
Reuters on fair use," and "grant[ed] partial summary judgment to Thomson Reuters on direct
copyright infringement for the headnotes in Appendix A" [source](https://www.govinfo.gov/content/pkg/USCOURTS-ded-1_20-cv-00613/pdf/USCOURTS-ded-1_20-cv-00613-5.pdf).

Two newsrooms independent of both parties carry the ruling firsthand. WIRED called it "the
first major AI copyright case in the United States" and reported that "Judge Bibas ruled
in Thomson Reuters' favor on the question of fair use" [source](https://www.wired.com/story/thomson-reuters-ai-copyright-lawsuit/). The Register described the
decision as one that "disallows fair use as a defense for training models on proprietary
data without permission," noting Ross "announced its shutdown on December 11, 2020,
following the Thomson Reuters lawsuit" [source](https://www.theregister.com/2025/02/12/thomson_reuters_wins_ai_copyright/). The grant is partial: the court left for a jury
whether some headnote copyrights had expired, and Ross has since appealed to the Third
Circuit (No. 25-2153) [source](https://www.govinfo.gov/content/pkg/USCOURTS-ded-1_20-cv-00613/pdf/USCOURTS-ded-1_20-cv-00613-5.pdf).

**Weakest load-bearing source, named where you meet it:** the courtroom facts above rest
on the court's own signed Memorandum Opinion (Tier 1) — the primary record, not a report
of it — so they are as strong as a source gets. The one claim that leans on a secondary is
the framing of this as the *"first major U.S. ruling"* to reject fair use for AI training
data: that superlative is a press characterization (WIRED, Tier 2), not a self-award in
the opinion, and the court itself confined its holding to Ross's *non-generative* search
tool and noted "the AI landscape is changing rapidly." Of the two independent secondaries,
The Register (a tech-trade outlet) is the weaker load than WIRED, and neither is the record
itself; the opinion is self-sufficient and the newsrooms only widen the corroboration
[source](https://www.wired.com/story/thomson-reuters-ai-copyright-lawsuit/).

**Verification status: `pending` (amber) — checker-graduated and publish-ready, awaiting a
HUMAN editorial sign-off. Not `verified`; no green badge is claimed here.**

## Path to green
This is a court-origin adjudication: the figures come from a signed, public federal
Memorandum Opinion, not any party's self-report, so no client-confirmation call is
required. The opinion is byte-identical to its GovInfo Wayback captures (CDX digest
match), and two newsrooms independent of both parties — WIRED (Condé Nast) and The
Register (Situation Publishing) — carry the ruling and the fair-use rejection firsthand,
each byte-tied to an organic Wayback capture. (Reuters News, though it also covered the
case, is Thomson Reuters's own corporate sibling and is deliberately not counted as an
independent source.) The remaining human step before a green badge is editorial: confirm
on the live D. Del. docket (No. 1:20-cv-613-SB) that the opinion is authentic and current,
and note the pending Third Circuit appeal — this story is scoped to the February 11, 2025
district ruling as reported.

## Related case files
- [Automattic (Tumblr / WordPress) AI-training opt-out](/stories/automattic-tumblr-wordpress-ai-training-opt-out-2024/) — the platform-consent counterpart: where Ross litigated after copying training data, Automattic scrambled to add an opt-out once users learned their content was being sold to AI trainers. The same "who may use this as training data" question, fought on the terms-of-service front.
- [Adobe terms-of-use AI-training clarification](/stories/adobe-terms-of-use-ai-training-clarification-added-after-backlash-2024/) — another training-data-rights flashpoint, showing how a vendor's own ToS language on AI training becomes contested exactly because rulings like *Ross* make that language legally load-bearing.
- [X suspends EU user-data training for Grok](/stories/x-suspends-eu-user-data-training-for-grok-after-irish-dpc-high-court-action-2024/) — a regulator-driven halt to using content as AI training data, the enforcement-side analogue to *Ross*'s copyright-side holding on the "data to train AIs" market.
- [WeTransfer AI-training clause removed](/stories/wetransfer-ai-training-clause-terms-removed-after-backlash-2025/) — a service walking back a clause that would have let it train on user files, illustrating the derivative "data to train AIs" market the *Ross* court found copyright-protected.

## Sources
1. **Tier 1 (primary).** U.S. District Court for the District of Delaware — *Thomson
   Reuters Enterprise Centre GmbH & West Publishing Corp. v. Ross Intelligence Inc.*, No.
   1:20-cv-613-SB, Memorandum Opinion (D.I. 770), filed February 11, 2025 (Bibas, Circuit
   Judge, sitting by designation); reported at 765 F. Supp. 3d 382. Official GovInfo PDF:
   https://www.govinfo.gov/content/pkg/USCOURTS-ded-1_20-cv-00613/pdf/USCOURTS-ded-1_20-cv-00613-5.pdf
   (byte-tied — local `sources/captures/ross-opinion-govinfo.pdf` sha1-b32
   JIDAU7I26HNP5A43JEZAIBWXYZZSG5OS, 286,902 B — to Wayback CDX digest across two organic
   captures of the official URL, 20250607072240 and 20250723213725, both status 200).
2. **Tier 2 (strong secondary, independent of both parties).** WIRED (Kate Knibbs),
   "Thomson Reuters Wins First Major AI Copyright Case in the US," February 11, 2025.
   https://www.wired.com/story/thomson-reuters-ai-copyright-lawsuit/
   (byte-tied — local `sources/captures/wired-raw.bin` sha1-b32
   RY7CAHKDJQULAFMKMPGCMNKHWJSXU3OR — to organic Wayback capture 20250211205714, status 200).
3. **Tier 2 (strong secondary, independent of both parties).** The Register (Thomas
   Claburn), "A win at last: Big-time blow to AI world in training data copyright scrap,"
   February 12, 2025.
   https://www.theregister.com/2025/02/12/thomson_reuters_wins_ai_copyright/
   (byte-tied — local `sources/captures/register-wb.bin` sha1-b32
   5ZN45V5AUOXBSDBHUA5T5Y2BY5N7XALG — to organic Wayback capture 20250212065402, status 200).

## How this was verified
- **Method:** Tier-1 federal Memorandum Opinion (D.I. 770), retrieved from the official
  GovInfo host and byte-tied to its organic Wayback captures via CDX digest; every quoted
  courtroom fact (the 2,243-headnote infringement finding, the grant of partial summary
  judgment, the revision of the 2023 ruling, the non-transformative and factor-four
  findings, the "data to train legal AIs" derivative market, the factor tally, and the
  grant of summary judgment on fair use) is quoted verbatim from that opinion. Two newsrooms
  independent of both parties — WIRED (Condé Nast) and The Register (Situation Publishing),
  each byte-tied to an organic Wayback capture — carry the ruling and the fair-use rejection
  firsthand. Reuters News is Thomson Reuters's corporate sibling and is deliberately NOT
  counted as an independent source.
- **Quote check:** all opinion and newsroom quotes above re-bound this session against the
  local `sources/captures/` files (whitespace- and curly-quote-normalized) — 12/12 opinion
  quotes and 6/6 newsroom quotes matched.
- **Date verified:** checker round 1 (confidence 1.000, two_independent on both critical
  claims; graduated to `pending`).
- **What "verified" (green) would still require:** a human editor confirming on the live
  D. Del. docket (No. 1:20-cv-613-SB) that the opinion is authentic and current, and that
  the pending Third Circuit appeal (No. 25-2153) has not reversed the district holding. No
  green badge is claimed here; this page is `pending` (amber).