Anthropic Will Pay $1.5B to Writers, Publishers Affected by Piracy

"At what point does training an AI on previously published works cross the line into piracy? A number of high-profile lawsuits are seeking to answer this very question. As Alex Reisner wrote in The Atlantic earlier this year, these lawsuits raise the question of whether or not "AI companies had trained large language models using authors' work without consent or compensation." This week, one of those cases, Bartz v Anthropic, concluded in a $1.5 billion settlement."

"In a settlement described by Cade Metz of The New York Times as "the largest payout in the history of U.S. copyright cases," AI company Anthropic agreed to pay this sum, which will involve the authors of 500,000 books each receiving $3,000. At issue in the case, Metz reports, wasn't necessarily the act of using these books for training purposes, which could be considered fair use. Instead, U.S. District Court judge William Alsup took issue with Anthropic using books that it knew had been pirated."

AI company Anthropic settled for $1.5 billion after using pirated books in its training datasets. The settlement allocates payments that would provide roughly 500,000 registered authors with $3,000 each. The court emphasized Anthropic's use of books it knew were pirated rather than the broader question of whether training on published works can be fair use. The Authors Guild identified seven million pirated books in Anthropic's databases, of which about 500,000 were actually used for training. Payment eligibility depends on U.S. Copyright Office registration and timing, and a final affected-works list will be submitted on Oct. 10.

Read at InsideHook