A recent ruling by a US court determined that using copyrighted material without permission for training AI does not qualify as 'fair use,' siding with copyright holders in the case between Thomson Reuters and Ross Intelligence.

Judge Stephanos Bibas highlighted that Ross's actions could harm the market value of Westlaw's offerings, as Ross aimed to compete directly with them.

In an effort to bolster its AI training, Ross acquired 25,000 Bulk Memos from LegalEase, which included questions and answers derived from Westlaw headnotes.

This ruling may prompt developers to explore alternative strategies for training AI models, such as utilizing synthetic datasets or user-generated content with explicit permissions.

The decision could lead to a shift in how AI developers approach training their models, emphasizing the need for proper licensing of content.

Judge Bibas reaffirmed the copyright status of Westlaw's headnotes, asserting that they possess originality due to the editorial judgment involved in their creation.

The court underscored that the market effect is a crucial factor in determining fair use, placing the burden of proof on Ross to show that its actions would not impact these markets.

The court found that both the headnotes and their key numbering system are original, with a significant number being copied and substantially similar to the originals.

Legal experts have noted that the ruling leans towards limiting fair use, although its implications may vary across different cases.

This ruling is significant as it opens a pathway for plaintiffs to challenge tech companies' fair-use defenses, despite the hurdles they may still face.

Judge Bibas acknowledged the public interest in accessing legal information but clarified that there is no right to Thomson Reuters's proprietary editorial content.

The distinction between generative and non-generative AI models may play a crucial role in future legal determinations, as generative models process data differently than non-generative ones.