A US federal judge allows the Kadrey v. Meta case, in which authors allege Meta violated IP rights by using their books to train Llama models, to move forward
A federal judge is allowing an AI-related copyright lawsuit against Meta to move forward, although he dismissed part of the suit.
Context & Ripple Effects
This ruling keeps a partially narrowed author challenge alive after an earlier dismissal of most claims in a related author suit against Meta. It makes the alleged use of books in Llama training a live merits question rather than ending the dispute at the pleading stage.
The case later became a major courtroom test of AI-training copyright claims, before a later decision found Meta's book-training use fair use on the arguments presented. That sequence underscores how procedural survival and a final copyright outcome can diverge.
First-order effects
- Meta must continue defending the surviving copyright claims over Llama's training data, while the authors retain a path to press their allegations despite part of the suit being dismissed.
- The decision preserves judicial scrutiny of the alleged book use without yet deciding whether that use infringed copyright.
Second-order effects
- The case gives other model developers and rightsholders a closely watched litigation vehicle for testing how training-data allegations can be pleaded and contested.
- Because the dispute proceeds on a partial claim set, parties in comparable cases are likely to focus more narrowly on provenance and the specific acts alleged, rather than treating model training as a single undifferentiated issue.
Third-order effects
- If courts continue separating the legality of training from the sourcing and retention of copyrighted material, AI copyright risk may turn increasingly on data-governance practices as much as model outputs.
- The later fair-use ruling in the same Meta book-training dispute suggests that surviving dismissal does not establish a durable liability standard; precedent will remain fact- and argument-dependent across training-data cases.
The trend: AI copyright litigation is moving from broad challenges to model training toward granular tests of data sourcing, copying, storage, and fair-use defenses.