AIAcademy · AIAcademy · 2026-05-16
Authors Guild — Bartz settlement explainer
The largest US copyright settlement on record. Anthropic agreed to pay $1.5 billion to a certified class of authors whose books appeared in pirated training corpora. The legal distinction Judge William Alsup carved out is the part worth studying — because it will be quoted in every LLM-training-data case that comes after.
The split Alsup made. Training on books lawfully acquired — bought, scanned from libraries with permission, licensed — is fair use. Training on books acquired through piracy — the LibGen and Books3 corpora that circulated in the early-2020s research community — generates damages. The Authors Alliance FAQ walks through the mechanics: the transformative-use analysis applied to the training step itself, but the underlying copy-acquisition step was a separate, infringing act.
What the settlement actually does. Anthropic pays roughly $3,000 per work in the class, with the precise figure depending on how many books are claimed. Authors had until early 2026 to opt in; the Authors Guild explainer details the claim form and the proof requirements. Anthropic does not admit liability and is not required to delete model weights or retrain.
What it implies for the next cases. The pending docket is substantial — see the AI Lawsuit Tracker for the running list. Kadrey v Meta, the New York Times v OpenAI matters, Tremblay, Silverman, Concord v Anthropic (the music-publishing parallel), and the Getty v Stability image cases all sit downstream of Alsup's reasoning.