U.S. Court Ruling Puts AI Training Practices Under The Microscope

A recent U.S. federal court ruling in Thomson Reuters Enterprise Centre GmbH v. Ross Intelligence Inc. determined that training artificial intelligence models on copyrighted materials does not fall under fair use protections—at least in the context of non-generative AI.

U.S. Circuit Judge Stephanos Bibas, presiding in the District of Delaware, issued the decision, reinforcing the legal rights of content creators and raising questions about how AI companies source and utilize training data. This ruling could have significant implications for copyright law and AI development, though its scope is limited to non-generative AI tools.

The Court’s Rationale: Copyright and AI Development

The case centered on whether Ross Intelligence’s use of copyrighted materials from Thomson Reuters’ Westlaw database to train its non-generative AI legal research tool qualified as fair use, a legal doctrine that allows limited use of protected works without permission for purposes such as commentary, criticism or research.

In its decision, the court found that:

• The use of copyrighted data was not transformative, as Ross’s tool aimed to create a competing legal research product rather than adding new expression, meaning or value to the original works. While Ross’s tool did not directly reproduce Thomson Reuters’ headnotes in its outputs, it relied on them as training data to build its platform.
• The commercial nature of Ross’s tool meant that it directly benefited from using copyrighted materials without compensating their creators.
• Ross’s actions harmed Thomson Reuters’ market for legal research platforms and its potential market for licensing data for AI training—a key factor in rejecting Ross’s fair use defense.

Judge Bibas emphasized that unauthorized use of copyrighted works to train AI systems does not automatically constitute fair use simply because outputs differ from the original material. He also warned that allowing unrestricted scraping of copyrighted content would undermine intellectual property laws.

iPhone SE 4: Apple Confirms New Product Launch Date Is Almost Here

Here’s Where Trump’s Government Layoffs Are—As 200,000 Recent Hires Could Be Affected

Implications for the AI Industry

This ruling could reshape how AI developers approach training models, particularly non-generative ones. Companies that have relied on scraping publicly available content may now face heightened legal risks, prompting a shift toward structured licensing agreements with content creators. This could lead to an industry-wide movement toward negotiating deals with publishers, artists, and authors to access high-quality data legally.

However, this decision is fact-specific and does not address whether training generative AI models on copyrighted materials constitutes fair use—a question likely to be addressed in future cases. For now, developers may need to explore alternative strategies such as creating synthetic datasets or relying on user-generated content with explicit permissions.

For content creators such as writers, musicians and visual artists who have long raised concerns about unauthorized use of their work by AI systems, this ruling represents a step toward greater control over how their creations are used in the digital age—providing stronger legal grounds to demand compensation or prevent misuse of their material.