US appeals court: ROSS training AI on Westlaw headnotes was not fair use

The US Court of Appeals for the Third Circuit upheld the ruling that ROSS Intelligence's copying of Westlaw headnotes to train an AI tool was not fair use.

FUFerhat UğurluYazSis TeamPublished: September 30, 20262 min read
In short

The US Court of Appeals for the Third Circuit affirmed a lower-court ruling in Thomson Reuters' case against legal research company ROSS Intelligence. According to IPWatchdog on 30 September 2026, the case covers more than 2,000 Westlaw headnotes, and the court found that ROSS's use of them for an AI tool was not fair use.

According to IPWatchdog on 30 September 2026, the US Court of Appeals for the Third Circuit affirmed the Delaware district court's ruling in the copyright case between Thomson Reuters and ROSS Intelligence. The decision was issued under seal; the court asked the parties to propose redactions within 10 days and said a suitably redacted version would be released afterwards.

  • The appeals court upheld partial summary judgment for Thomson Reuters on its claims that ROSS infringed more than 2,000 Westlaw headnotes, and upheld the denial of ROSS's summary judgment bid based on fair use.
  • In the district court, Judge Stephanos Bibas found in a revised February 2025 opinion that ROSS built a competing product and that its use was not transformative.
  • LawSites notes that ROSS built a non-generative legal search tool, that the company has since shut down, and that the copying took place before generative AI became widespread.
  • According to an update from LawSites, ROSS says it will ask the US Supreme Court to review the ruling.

Why it matters

IPWatchdog and LawSites describe the decision as the first US appellate ruling on fair use in AI training. LawSites adds that its effect on cases against OpenAI, Anthropic and Meta depends on the full opinion's reasoning, and that the specific facts of the case may limit it. The ruling creates no direct obligation for companies in Türkiye, but it could suggest that training a model on a competitor's database may carry serious legal risk. Companies that train models on their own data or buy datasets from third parties may want to document where the data comes from and under which licence terms.

Sources

FU
Ferhat Uğurlu · YazSis Team

Builds enterprise software, AI and digital transformation products together with the YazSis team.