Third Circuit Ruling: AI Training Data Is Not Fair Use, Reshaping Creator Rights

Seed story: "Court Finds That Using Copyrighted Material as AI Training Data Is Not Fair Use" (Futurism) · search original Written from facts verified across 3 news report(s) — original explainer, not a copy or translation. Sources listed at the end.

A recent Third Circuit decision has clarified that using copyrighted material to train AI models does not qualify as fair use, a ruling that significantly raises the legal stakes for creators whose work is incorporated into generative systems. This determination, which affirmed a loss for a defunct legal AI startup, suggests that liability for training data usage is shifting from a gray area to a clearer infringement risk, potentially reshaping how freelancers and rights holders approach licensing and protection in the AI era.

The Ruling: Why AI Training Failed the Fair Use Test

On September 29, 2026, the Third Circuit Court of Appeals affirmed a ruling in favor of Thomson Reuters against Ross Intelligence. The court explicitly rejected the fair use defense for using Westlaw headnotes to train an AI search engine. This decision marks a significant legal boundary for data-driven legal tech.

The court found that Ross’s use was not transformative because it directly competed with Thomson Reuters’ offerings. Rather than generating new content, the model was designed to improve search results for existing court decisions. Key factors in this determination include:

The use of "Bulk Memos" created from Westlaw headnotes by a third party, LegalEase.
The direct competitive nature of the AI tool against the original database.
The lack of generative transformation in the output.

For creators, this signals that merely repurposing existing data for search optimization does not constitute a new, transformative work. It reinforces that direct competition with the source material undermines fair use claims, potentially affecting how legal and professional data is licensed.

Direct Competition and the 'Transformative' Myth

The Direct Competition Problem

Judge Bibas’s analysis hinges on a critical distinction: Ross’s AI did not create new artistic or informational value. Instead, it functioned as a direct competitor to Thomson Reuters by improving search results for existing court decisions. Because the tool replicated the core utility of Westlaw rather than generating novel content, the court rejected the "transformative" argument. This stands in sharp contrast to generative AI models that synthesize new outputs, which often face different legal hurdles.

The ruling clarifies that "transformation" does not merely mean using data in a new technical process. If the end product serves the same market purpose as the original work, fair use is unlikely to apply. For creators, this distinction is vital:

Search vs. Generation: Tools that index or retrieve existing content face higher risk than those creating new derivatives.
Market Harm: Direct competition with the rights holder’s primary offering weighs heavily against fair use.
Technical Neutrality: Using AI as a backend mechanism does not automatically shield the output from copyright claims.

This precedent suggests that creators licensing data for retrieval-based AI systems must negotiate explicit permissions, as implied licenses may not cover direct market substitution.

The Broader Legal Landscape: From Anthropic to Meta

The Third Circuit’s September 2026 affirmation in the Thomson Reuters case signals a decisive shift toward stricter liability for AI developers. This ruling aligns with the broader trend seen in the $1.5 billion settlement of the Bartz v. Anthropic lawsuit, where a class payment distribution plan was approved on September 15, 2026. These developments suggest that courts are increasingly rejecting the notion that training on copyrighted material constitutes a protected, transformative act.

Key recent developments include:

Anthropic Settlement: A $1.5 billion resolution approved by Judge Martínez-Olguín, establishing a significant financial precedent for class actions.
Meta Discovery: Judge Chhabria granted additional discovery on a contributory-infringement claim in the Kadrey v. Meta case on August 11, 2026.
International Watch: The Robert Kneschke v. LAION e.V. case remains pending before the Federal Court of Justice following a September 3, 2026 hearing.

For creators, this landscape implies that voluntary licensing is becoming the primary legal safeguard. As litigation costs rise and settlements grow, relying on implied permissions is no longer a viable strategy for protecting intellectual property rights.

Implications for Creator Contracts and Licensing

Shifting the Burden to Developers

The Third Circuit’s affirmation on September 29, 2026, fundamentally alters the economic landscape for creators. By rejecting the fair use defense, the ruling forces AI developers to secure explicit licenses rather than relying on implied permissions. This shift moves the burden of compliance directly onto technology companies, requiring them to negotiate and pay for the rights to use copyrighted material for training purposes.

For creators, this creates a new leverage point in contract negotiations. Instead of passively watching their work be scraped, professionals can now demand:

Explicit licensing agreements that define usage scope.
Clear compensation structures for data inclusion.
Retention of control over how their work is represented.

As seen in the Ross Intelligence case, where the company shut down in 2021 due to litigation costs, the financial risk of unauthorized use now falls on the developer. This encourages a market where creators are compensated partners rather than incidental data sources, ensuring that rights and payments are explicitly negotiated before any training begins.

Strategic Actions for Independent Professionals

Protecting Your Digital Assets

With the Third Circuit affirming that AI training on copyrighted material is not fair use, independent creators must proactively audit their digital footprints. Since the court rejected the "transformative" defense for direct competition, your work is now a distinct asset rather than free fuel.

Key steps include:

Audit visibility: Review where your content is indexed to understand potential exposure.
Update contracts: Ensure licensing agreements explicitly prohibit AI training use.
Monitor usage: Track if your work appears in AI outputs or datasets.

This shift means creators can enforce rights more aggressively. As seen in the Bartz v. Anthropic settlement, legal leverage is growing. By documenting your IP and updating terms, you position yourself to claim compensation if your work is used without permission, turning legal precedent into tangible financial protection.

FAQ

Did the Third Circuit rule that using copyrighted data for AI training is fair use?

Yes, the Third Circuit affirmed that Ross Intelligence's use of Thomson Reuters' Westlaw headnotes to train its AI search engine was not fair use. The court determined the use was not transformative because it directly competed with Thomson Reuters' existing offerings rather than generating new content.

What happened to Ross Intelligence after the lawsuit was filed?

Ross Intelligence shut down in 2021 shortly after the lawsuit was filed, citing financial strains caused by the litigation. The company had previously hired a third party named LegalEase to provide training data consisting of 'Bulk Memos' created from Westlaw headnotes.

How does the Thomson Reuters ruling compare to other recent AI copyright cases?

While the Third Circuit ruled against Ross Intelligence, the Bartz v. Anthropic lawsuit was settled for $1.5 billion with a class payment plan approved in September 2026. Additionally, the Kadrey v. Meta Platforms case is ongoing, with Judge Chhabria granting Meta additional discovery on a contributory-infringement claim in August 2026.