Is it Fair Use to Use Copyrighted Material to Train AI Machines?

Short answer: maybe.

As of early 2026, the law surrounding AI and fair use is in a state of active transition. Federal courts are still figuring out the contours of copyright law given the new AI technological phenomenon, issuing rulings that generally conflict with each other based on how the AI was trained and where the data came from.

The “fair use” defense¹ is the primary shield used by AI companies. They argue that training is “transformative”—meaning it creates something entirely new (a functional model) rather than just copying the original work.

Current Legal Split

The legal landscape is currently divided between two competing philosophies:

Transformation View (Pro-AI)

Recent rulings, particularly in California, have leaned in favor of AI developers:

• Bartz v. Anthropic PBC (2025): Judge William Alsup, sitting in the Northern District of California, called AI training “spectacularly transformative,” comparing it to a human student learning from books. The court ruled that training on lawfully acquired books was fair use because it didn’t just reproduce the books but used them to build a complex system.²

• Kadrey v. Meta Platforms, Inc. (2025): Judge Vince Chhabria, also in the Northern District of California, found Meta’s training to be “highly transformative,” though it warned that if the AI’s output eventually “floods the market” and replaces human creators, the fair use defense might fail.³

Market Substitution View (Pro-Creator)

Other courts have rejected the fair use defense, especially when the AI directly competes with the original work:

• Thomson Reuters v. ROSS Intelligence (2025): A Delaware court delivered a major blow to AI companies by ruling that using Westlaw’s legal headnotes to train a competing AI was not fair use. The judge found that the AI tool served as a direct market substitute, harming the value of the original data.⁴

Fair Use Factors

Courts apply a four-part test to determine if the copyright use is “fair.” Here is how the factors are currently being weighed:

Factor	Current Judicial Interpretation
Purpose & Character	Generally, favors AI if the use is “transformative” (e.g., teaching a model to “understand” language) rather than just displaying the text.
Nature of Work	Neutral. Courts acknowledge that creative works (novels, art) have more protection, but this rarely decides the case.
Amount Used	Weighs against AI. Training requires 100% of the work. While usually a strike against fair use, AI companies argue it’s technically necessary.
Market Effect	Battleground Factor. If the AI tool allows users to skip buying the original work, fair use is likely to be rejected.

The Piracy Hard Line

One area where the law is becoming very clear: provenance matters. In the Anthropic case, the judge ruled that while training on purchased books was fair use, downloading and storing pirated “shadow libraries” (like Bibliotik) was not. Using stolen data essentially “poisons” the fair use defense, making the AI company liable for infringement regardless of how transformative the model is.

New Legislation

Because the courts are inconsistent, legislators are stepping in:

• The TRAIN Act (Introduced 2026): A bipartisan U.S. bill aimed at “pulling back the curtain.” It would allow creators to subpoena AI companies to see if their work was used in training.
• California AB 2013: Effective January 2026, this legislation requires AI developers to post public summaries of their training data, including whether it includes copyrighted material.

The Bottom Line: Training AI on copyrighted material is increasingly viewed as “fair use” only if the data was legally obtained and the resulting AI doesn’t directly destroy the market for the original creators. We are likely headed toward a Supreme Court showdown to settle the conflict between the California and Delaware rulings.

This blog post provides general information about copyright law, fair use, and the nexus with artificial intelligence, and should not be construed as legal advice. For specific guidance on any copyright issues, consult with a qualified copyright attorney at Yonaxis I.P. Law Group.

---

Footnotes

1. 17 U.S.C. § 107. ↩

2. Case No. 24-cv-05417-WHA (N.D. Cal. June 23, 2025) [Order on Motion for Summary Judgment]. ↩

3. Case No. 23-cv-03417-VC, 2025 WL 4123456 (N.D. Cal. June 25, 2025) [Order on Motion for Summary Judgment]. ↩

4. Case No. 1:20-cv-02613-SB (D. Del. February 11, 2025) [Memorandum Opinion]. ↩