Thomson Reuters v. ROSS — Third Circuit Oral Argument June 11 #
- Third Circuit sets oral argument for June 11 in 1st appeal of decision on fair use in AI training (Chat GPT Is Eating the World, 2026-04-14) — The first AI training data fair-use case to reach circuit court level. Background: Judge Bibas (Delaware) reversed his own 2023 finding and held in 2025 that Westlaw headnotes used to train ROSS were not fair use. Two hard questions before the Third Circuit: (1) whether the use was transformative; (2) whether Westlaw headnotes meet the originality threshold. Both parties filed supplemental briefs on ASTM v. UpCodes, disagreeing on what it means for this case.
- Thomson Reuters, ROSS Intelligence disagree on meaning of Third Circuit’s ASTM v. UpCodes in supplemental briefs (Chat GPT Is Eating the World, 2026-05-12) — Supplemental brief battle: Thomson Reuters argues ASTM confirms copyright protection for curated works; ROSS argues ASTM limits protection to literal text, not functional assemblage. The disagreement is about the scope of copyright in AI-processable data structures — a foundational question for the entire industry.
Discovery Expands — OpenAI Must Produce 20 Million ChatGPT Logs #
- OpenAI Must Turn Over 20 Million ChatGPT Logs, Judge Affirms (Bloomberg Law) — Judge Stein (SDNY) affirmed January 5, 2026 that de-identified ChatGPT logs are discoverable even when they don’t contain plaintiffs’ works — because they bear on OpenAI’s fair use defence. Users voluntarily submitted conversations, so privacy interests don’t override discovery. Structural implication: AI model outputs are now routinely evidence in copyright litigation.
Legislation — Bipartisan TRAIN Act #
- Dean, Moran Introduce Bipartisan Bill to Protect Creators from Unauthorized AI Training (Congresswoman Dean, 2026-01-22) — H.R. 7209 (TRAIN Act): adds an administrative subpoena process to the Copyright Act, allowing copyright owners to compel AI developers to disclose training data contents. Senate cosponsors: Welch (D-VT), Blackburn (R-TN), Schiff (D-CA), Hawley (R-MO). Bipartisan backing signals this has traction even in a Congress that has otherwise stalled on AI legislation.
Cross-links #
- AI Impact on Society Colorado SB 26-189 regulatory retreat is simultaneous with copyright law tightening through courts — legislatures are easing while courts apply existing law independently. The accountability mechanisms are inverting.
- Open vs Closed AI Ecosystems The TRAIN Act’s subpoena mechanism creates discovery asymmetry: closed labs are easier to subpoena than open-weight model developers who distributed weights widely. This is a structural compliance advantage for open-weight approaches in avoiding IP liability.
- Quality signal: Thomson Reuters v. ROSS is now the most important AI copyright case in any court. It combines: (1) the originality question (are curated AI-processable data structures copyrightable?); (2) the training use question (is AI training transformative fair use?); (3) the first circuit-level ruling on either. June 11 is the inflection date.
- Emerging theme: AI outputs (ChatGPT logs) are now discoverable in copyright litigation. This creates a new disclosure surface — anything a model says can be used to demonstrate what it absorbed from training data.