01
Factual background & dispute
- Multiple authors alleged that their copyrighted books were incorporated into training datasets for Meta's Llama models.
- The parties submitted evidence regarding training purposes, output market substitution, primary book markets, and potential licensing markets.
- The court confined its determination strictly to the specific evidence submitted by the parties at that stage.
02
Core issues & judicial focus
- How courts assess the transformative character of large language model training
- How plaintiffs must prove the market impact of AI-generated content on original literary works
- Whether an AI training licensing market qualifies as a protectable market under copyright law
03
Judicial finding & holding
- On this record, the court found transformative training use and insufficient evidence of relevant market harm from the 13 authors, granting fair-use summary judgment on training-copying claims.
- ECF700 of 25 March 2026 allowed amendments to contributory and distribution claims and the proposed class definition, while denying class discovery at that time.
- Remaining issues continued, with further discovery orders in September. The decision was not a defeat of a certified class or a finding that all acquisition of training material was lawful.
Dates & procedural record
- ECF598 entered summary judgment on 13 authors’ training-copying claims.
- ECF700 allowed specified amendments; other disputes continued.
- The case directory records a further discovery order.
04
Practical risk implications
01Rights holders must assemble concrete evidence showing dataset inclusion, infringing model outputs, and direct market substitution.
02AI developers should document training purposes, data processing workflows, and output filtering controls.
03U.S. courts apply varying degrees of scrutiny to market harm evidence in AI copyright litigation.