Meaghan Kent, an intellectual property partner at Venable LLP, discusses recent court rulings on whether artificial intelligence training models are free to draw on copyrighted content.

There’s a substantial amount of litigation underway on the issue of copyrighted content being incorporated into generative AI models. Kent discusses three recent federal court rulings, in the cases of Thomson Reuters v. ROSS Intelligence, Claude.ai v. Anthropic, and Kadrey v. Meta Platforms.

The Reuters case involved the use of AI to summarize case headnotes created by Thomson Reuters’ West Publishing Corp. A judge found that the action amounted to establishment of a competitive product, and was therefore not protected by fair use law under the Copyright Act of 1976.

The second case involved a group of authors suing Anthropic for drawing on their works for training its Claude.ai assistant. The case revolved around the AI model collecting data from two separate sources: existing works that were considered to be pirated versions of protected content, and copies of authorized works that were purchased and then scanned. A judge ruled that the pirated works were not subject to fair use law, but that the principle did apply to the purchasing and scanning because those actions weren’t likely to cause “dilution” of demand for the authors’ works in the marketplace and, further, that the resulting content was sufficiently transformative in nature to avoid copyright claims. (Editor’s note: Anthropic has agreed to a settlement of $1.5 billion to the authors and publishers who brought the suit. According to the New York Times, it’s the largest payout in the history of U.S. copyright disputes.)

In the third case, involving the training of Meta’s AI model known as LLaMA, a judge rejected plaintiffs’ copyright-infringement claims, while indicating that he would have considered the “market dilution” argument, had the plaintiffs submitted evidence of harm.

The issue of AI and fair use is far from settled, and Kent says intellectual property owners and AI developers need to keep abreast of fast-changing developments. A likely outcome in the short term, he added, is copyright owners agreeing to license their content to multiple AI providers, a trend that is already well underway.