The big picture. The final approval (fairness) hearing in Bartz v. Anthropic was scheduled for May 14, 2026 in the U.S. District Court for the Northern District of California. The $1.5 billion settlement — the largest copyright settlement in U.S. history — covers the class action’s claims that Anthropic illegally copied books to train Claude.

The underlying ruling.

  • Judge William Alsup ruled on summary judgment that using legally acquired books to train AI is fair use.
  • Alsup separately denied Anthropic’s request for summary judgment on its use of pirated copies, finding piracy was not fair use.
  • The case settled after that bifurcated ruling.

Settlement terms.

  • Total: $1.5 billion.
  • Per-work payout: approximately $3,000 per copyrighted work, according to the Authors Guild.

Why it matters. Judge Alsup’s split ruling is now the operative line on AI copyright in the Northern District of California, the venue that houses the largest concentration of AI copyright litigation. The line: “training on legally acquired books = fair use; training on pirated copies = not fair use” is codified by an industry-shaping settlement, even if the appellate question on the fair-use side remains open elsewhere.

Adjacent cases still active.

  • Kadrey v. Meta. The court granted partial dismissal on fair use for training, but Meta still faces active claims related to alleged reproduction of pirated works during “seeding” in the torrenting process.
  • Thaler v. Perlmutter. Supreme Court denied certiorari on March 2, 2026, confirming that AI-generated material requires human authorship to be copyrightable.

What’s next. The post-settlement question is whether other labs facing similar actions follow Anthropic’s settlement structure, contest the split-ruling framework on appeal, or restructure their training-data acquisition to remove pirated sources from the question entirely.

Sources

Sources