Robert Ambrogi Blog
The long-running copyright litigation between legal research giant Thomson Reuters and now-shuttered legal research startup ROSS Intelligence took another interesting turn this week, following the 3rd U.S. Circuit Court of Appeals’ order that the parties file supplemental briefs addressing the impact of its recent ruling in American Society for Testing & Materials v. UpCodes, Inc. — a case I covered here in April.
Both sides filed their supplemental briefs on May 11, and, not surprisingly, each argues that UpCodes supports its position. What is notable, though, is how different are their readings of the same opinion — and how directly the briefs crystallize the central dispute at the heart of this appeal.
Find all of my coverage of this litigation here.
The filings came in response to an April 27 text order from the court’s clerk directing the parties to file supplemental briefing “addressing what impact, if any” the court’s UpCodes decision has on the appeal.
As I wrote in April, the UpCodes decision was a significant fair use ruling in which a unanimous 3rd Circuit panel held that UpCodes’ publication of technical building standards incorporated by reference into law likely constitutes fair use.
The court found UpCodes’ use transformative because its purpose — conveying what the law actually is — differed fundamentally from ASTM’s purpose in publishing those same standards.
ROSS: UpCodes Requires Reversal
In its supplemental brief, ROSS, which is represented by law firms White & Case and Pillsbury Winthrop Shaw Pittman, argues that UpCodes is essentially controlling and demands summary reversal of the district court’s ruling against it on fair use.
When ROSS initially petitioned for this interlocutory appeal, the question of whether AI innovators could copy available legal materials to build a new research platform was described as “novel and difficult,” the brief argues. Then, last month’s unanimous UpCodes panel answered that question: Yes, fair use permits it.
ROSS draws parallels between its situation and that of UpCodes. West, like the UpCodes plaintiff, is a recognized industry leader whose users purchase subscriptions to access copyrighted materials. ROSS, like the UpCodes defendant, is a for-profit startup that built a legal research platform using “copyrighted” content (the word is in quotes in the brief) without securing a license, and both employed freemium business models offering AI-powered tools.
ROSS goes factor by factor through the fair use analysis, arguing that UpCodes requires reversal on the first and fourth factors, where the district court ruled against it, and affirmance on the second and third, where the district court found for ROSS.
On the critical first factor — the purpose and character of the use — ROSS argues the district court made two errors that UpCodes now exposes. First, the district court analyzed ROSS’s subjective purpose (it found ROSS “took the headnotes to make it easier to develop a competing legal research tool”), whereas UpCodes clarifies that the first factor calls for an objective inquiry.
Second, the district court conflated ROSS’s purpose with the expressive similarity of the output, which UpCodes says is the wrong inquiry. ROSS’s objective use — training an AI model on the relationships among legal words, so that the headnotes would not appear in the final product at all — is, ROSS contends, at least as transformative as what UpCodes did, and arguably more so, because it “involved a change in medium.”
On the fourth factor — market harm — ROSS argues the district court made a categorical error by treating “legal research platforms” as the relevant market, rather than examining what West actually derives from the specific works at issue, its headnotes. Under UpCodes, ROSS says, the court should have looked at the market for headnotes specifically and the percentage of Westlaw revenues attributable to them.
ROSS also argues that West conceded it has no evidence of actual market harm — no documented lost subscribers, no lost revenue traceable to ROSS — and that UpCodes requires courts to examine the record closely rather than inferring harm from the mere fact of competition.
“In 2020, West should have responded to ROSS with innovation,” the brief concludes. “Instead, West chose litigation. Copyright rewards the former, not the latter.”
Thomson Reuters: UpCodes Changes Nothing
In its supplemental brief, Thomson Reuters, represented by Kirkland & Ellis, takes the opposite view: UpCodes does not move the needle here at all, and in fact demonstrates why ROSS’s conduct falls on the wrong side of the fair use line.
TR’s brief argues that the fundamental distinction in fair use doctrine is between commercial, substitutive uses (which are not protected) and uses serving a genuinely different purpose that do not supplant the original work’s markets. UpCodes, TR contends, is simply an application of that settled framework — and the factual differences between that case and this one point squarely against ROSS.
The key distinction, TR contends, is that UpCodes copied standards that had been legally incorporated into law and made them freely available, serving a genuine access-to-law purpose it could not achieve without copying those specific works. But ROSS had no such access-to-law purpose. The case law was already free. ROSS had it.
Further, unlike UpCodes — which copied only the law-incorporated versions of standards and made them available at no charge — ROSS was fully commercial, explicitly marketing its product as a direct Westlaw replacement and, according to TR, even bragging to prospects that it would “bring an end to Westlaw contracts.”
TR also invokes a California district court’s recent ruling in Bartz v. Anthropic PBC, which it quotes for the proposition that using a proprietary system for finding court opinions to create a competing AI tool for finding court opinions “was not transformative.”
On the second factor, TR argues that UpCodes actually underscores the distinction. The UpCodes works had “legal force” because they were incorporated into statutes. Westlaw headnotes are not law. They reflect TR’s attorney editors’ legal analysis and organization — proprietary editorial work that the Supreme Court recognized as copyrightable in Callaghan v. Myers back in 1888.
On the fourth factor, TR argues the evidentiary record here is far richer than in UpCodes, where market harm evidence was thin and many of the incorporated standards at issue were outdated.
By contrast, the record here, TR says, shows customers who switched from Westlaw to ROSS, ROSS’s CEO acknowledging that switch, and evidence of harm not just to the original Westlaw subscription market but to the emerging market for AI training data — a harm the UpCodes panel never had to address.
The Bottom Line
To some extent, the parties’ differing interpretations of the Upcodes ruling reflects the ambiguity in the ruling itself. The 3rd Circuit grounded its fair use analysis heavily in UpCodes’ specific purpose of disseminating the law, a purpose it found compelling and independent.
Whether that reasoning extends to AI training on editorial legal content, or whether it is restricted to incorporated-standards cases, is exactly the question the 3rd Circuit now has to answer in this appeal.
ROSS’s argument — that teaching a machine the language of the law is analogous to UpCodes’ dissemination of legal standards — is creative and timely, given how AI has developed during the pendency of this case.
TR’s counter — that ROSS’s purpose was simply to replicate what Westlaw does — has the advantage of following the district court’s reasoning and finding factual support in the record.
For now, the ball is back in the hands of the court.