Fair Use in 2025 – the courts draw new lines

August 29, 2025
AI & Digital Rights • Broadcaster Rights • Copyright Enforcement • Creator Protection • Media Law

Reviewed September 2026: This revised discussion distinguishes the June 2025 AI-training decisions from the later Anthropic settlement. It replaces the earlier forecast of a December trial.

The 2025 decisions in Bartz v. Anthropic and Kadrey v. Meta are important examples of fair-use analysis. They do not give every AI system permission to copy every work. The particular copying, purpose, and evidence matter.

Bartz: Training and the Library Were Separate Uses

In the June 23, 2025 order, Judge William Alsup held that the training use before him was fair use. He separately approved the conversion of purchased print books into digital replacements under the circumstances addressed in the order.

The court did not excuse downloading pirated books to build a permanent central library. This distinction is more precise than saying that any unauthorized source automatically defeats fair use for every later training use. Acquisition, retention, and training were analyzed separately.

Kadrey: The Evidentiary Record Mattered

Judge Vince Chhabria’s June 25, 2025 order granted Meta summary judgment on the training claim presented by thirteen authors. The opinion emphasized shortcomings in their arguments and evidence, including evidence of market dilution from model outputs.

The court expressly limited the decision to those plaintiffs and that record. It did not declare Meta’s training practices universally lawful. The two opinions also should not be flattened into a single rule about transformation or market harm.

The Anthropic Settlement Update

On July 20, 2026, the court granted final approval of the $1.5 billion class settlement in Bartz. That development supersedes the earlier trial forecast. Settlement approval is distinct from the June 2025 fair-use ruling; it is not a new blanket ruling on AI training.

What Creators and Publishers Can Take From This

Keep evidence of what work was copied, how it was obtained, and how the challenged use affects the work or its market. Distinguish a claim about acquiring a library from one about training a model or distributing an output.

A headline about an AI company winning or settling a case cannot answer those questions for a different dispute. For guidance on an archive or creative work being used for training, explore AI use of your content or archive or request a consultation.


Related Resources

Originally published by TALKERS on August 29, 2025. Adapted and updated for this Harrison Legal Group edition in September 2026.