The ruling has been closely watched. But the court went out of its way to say what the case was not—and to distinguish it from cases dealing with generative AI platforms. As the court put it, “this is no more than an ordinary copyright case.” Understanding that framing is the key to understanding what the decision means for your business.

What Happened

ROSS set out to build an AI-powered legal search engine that would answer plain-English legal questions by returning relevant passages from court opinions. To teach its system what a good response looked like, ROSS hired a vendor to prepare roughly 25,000 training memos. But the vendor’s writers used Westlaw headnotes—the short summaries of legal points that Westlaw editors write for each opinion—to create the training memos that ROSS’s AI program used to identify patterns regarding which passages from court opinions respond well to legal questions. The vendor did so using borrowed Westlaw credentials and accounts opened under misleading pretenses. ROSS openly marketed its service as an alternative to Westlaw, at comparable prices. ROSS argued that its reliance on the headnotes was “fair use.”

The court found that 2,243 of those headnotes were copied, that each reflected enough editorial judgment to be protected, and that ROSS’s use failed three of the four fair use factors. Most significantly, in this fair use case as in others, the court found that ROSS used the material for essentially the same purpose as Westlaw—helping users find relevant case law—and indeed to compete directly with it.

Not Generative AI, and Not an “AI Case”

Two features of the decision deserve emphasis. First, ROSS’s system was not generative AI. It did not write anything new; it simply retrieved existing passages from court opinions. The court expressly distinguished cases involving large language models, such as the litigation against OpenAI, noting that tools capable of generating original output raise different questions about whether training is “transformative.”

Second, the court declined to treat AI as a category deserving special rules. It rejected ROSS’s arguments that a ruling against it would halt AI development or harm national security, observing that developing AI does not provide “carte blanche” to disregard copyright law. Instead, the court applied the same principles it would apply to any copying: Was the material original? Was the use genuinely different in purpose? Did it harm the owner’s market?

Key Takeaways

Like many fair-use cases, the opinion is fact-specific. Still, it contains several observations worth keeping in mind inside and outside the AI context:

• Competing with the source is the riskiest posture. In weighing the fair use factors, the Third Circuit repeatedly emphasized that ROSS built a commercial substitute for Westlaw. Combined with the absence of any “transformation” in ROSS’s use of the Westlaw headnotes, that competitive purpose defeated fair use. It is not unusual in fair-use cases for the infringer’s competitive purpose to be the dominant—or even the decisive—factor. To that point, the court distinguished disputes involving generative AI laboratories, observing that “the evidence” in those cases indicated a higher degree of transformativeness and a lesser degree of direct competition.
• There is no special copyright law for AI. The Third Circuit went out of its way to emphasize that, even where “new technology” is involved, courts “regularly apply “the principles underlying our venerable copyright laws.” In this case, as in others, the ordinary fair-use factors dictated the outcome. AI is subject to the same law as older technologies.
• Vendors’ conduct can be attributed to you. The copying attributed to ROSS was performed by a contractor and its subcontractor, yet ROSS bore the consequences.
• How you obtain data matters. The court noted evidence that ROSS sought Westlaw access through borrowed credentials and accounts opened under misleading pretenses. Those facts precluded any finding that ROSS’s copying was done in “good faith.”
• Convenience is not a defense. ROSS could have built its training data from freely available court opinions. It chose headnotes because they were “easy.” The court held that “ease is not a justification for copying.”

Practical Considerations

For companies that develop AI, the decision is a reminder that the provenance of training data is a core legal issue, not a technical detail. Businesses need to know, with confidence, where their training data comes from and on what terms; whether their vendor and data-sourcing agreements address ownership, permitted uses, and responsibility for third-party claims; and whether any data was gathered in ways that could conflict with website terms of service or subscription agreements.

For companies that use AI tools, consider the other side of those same questions. Do your agreements with AI providers include meaningful protection if a tool proves to have been built on improperly sourced material? And for companies that own valuable content—databases, research, editorial products, or proprietary documents—the decision may strengthen the hand of rights holders seeking to license their material or to respond to unauthorized use.

Looking Ahead

The Third Circuit has supplied an important fair-use precedent, but the decision has clear limits. Questions regarding generative AI technology remain unresolved and are working their way through other courts. In the meantime, the clearest lesson is that familiar copyright principles apply with full force to new technology as well as old. And decisions made early in development—about data sources, vendors, and contracts—can carry consequences years later.

This alert is for informational purposes only and does not constitute legal advice.