Summary

June 24, 2025 — A federal judge ruled this week that Anthropic’s use of lawfully purchased books to train its Claude AI models qualifies as *fair use* under U.S. copyright law. However, unresolved claims related to pirated content remain headed to trial, leaving key legal questions open—and the ruling’s long-term impact uncertain.

What the Judge Decided

Transformative use allowed: U.S. District Judge William Alsup ruled that converting legally purchased books into a format suitable for training large language models (LLMs) is “quintessentially transformative.” Because the books were altered in purpose—from human reading to machine learning—the use was found to support the public interest in innovation and knowledge creation.

No fair use for pirated content: By contrast, Alsup rejected fair-use defenses for content acquired from shadow libraries. That part of the case will proceed to trial in December.

How Precedential Is This?

Although this is a district court ruling (not binding nationwide), its influence may be substantial:

First-of-its-kind interpretation: This is the first major federal decision explicitly holding that training an AI model on legally acquired copyrighted books can be protected under fair use. It provides a legal blueprint for other courts and tech companies navigating similar issues.

Guidance for pending cases: Dozens of lawsuits have been filed against OpenAI, Meta, Stability AI, and others for allegedly misusing copyrighted content. While those cases are still developing, this decision could serve as persuasive authority, particularly in jurisdictions where judges are weighing how transformative use doctrine applies to AI training.

Not a circuit-level or Supreme Court ruling: Importantly, this decision does not bind other courts. It’s only controlling in this particular case unless adopted by an appellate court (such as the Ninth Circuit) or ultimately affirmed by the U.S. Supreme Court. Other judges could interpret the scope of fair use in the AI context more narrowly—or more broadly.

Risk of reversal or limitation: If appealed, the Ninth Circuit could reverse or limit this ruling. Courts may diverge over what counts as “transformative,” particularly when the underlying works are used at massive scale, and when model outputs raise additional copyright concerns. A future ruling could find that even purchased materials, if used too comprehensively, exceed fair-use boundaries.

Implications for the AI Industry

Companies may lean on this framework: AI developers now have a clearer (though not foolproof) path for avoiding infringement—by ensuring training sets are composed of purchased or properly licensed materials and are used in a clearly transformative way.

Licensing pressure remains: The ruling doesn’t eliminate the economic incentive for rights holders to demand compensation. Expect continued calls for compulsory licensing schemes or collective bargaining frameworks, especially for visual, music, and audiovisual content.

Unanswered questions about model outputs: This decision focused on inputs—the use of copyrighted text in training. It did not address whether model-generated outputs that resemble or reproduce parts of those works could constitute infringement. That legal frontier remains open.

What Comes Next

– December 2025 trial: Anthropic still faces substantial liability for its alleged use of pirated books. Statutory damages could reach hundreds of millions, depending on findings at trial.

Appeals likely: Regardless of the outcome, an appeal to the Ninth Circuit appears almost certain. That will be the next major opportunity for precedent-setting at a higher level.

Regulatory developments: This case will also shape ongoing debates in Congress and international forums over copyright reform, particularly around transparency in AI training and model explainability.

Final Thoughts

Judge Alsup’s ruling is a meaningful step toward defining what fair use looks like in the age of generative AI. But it is just that: a step. The legal framework for AI training remains in flux, and this decision—while groundbreaking—could still be reined in, overturned, or distinguished in future cases. For now, it stands as a signal to developers: thoughtful sourcing matters, and courts are beginning to draw clearer lines between innovation and infringement.