Unsealed filings: OpenAI, Microsoft knew book piracy was illegal

Unsealed court filings in the Authors Guild's consolidated case against OpenAI and Microsoft include an internal 2019 message in which OpenAI staff acknowledged training GPT-3 on pirated books. "We trained GPT-3 on pirated stuff! No sharing that!" one note reads, according to briefs filed September 17-18 in the Southern District of New York and reported by Publishers Weekly.

The documents, in Alter v. OpenAI and Microsoft (25-md-03143), show OpenAI policy director Jack Clark writing in May 2020 that "the better we do on GPT-X, the more worried genre fiction authors will become about us substituting for them on Amazon." Clark added, per the briefs, that "we'll likely ignore their concerns and release anyway." OpenAI research director Dario Amodei separately described LibGen, the shadow-library source of many of the books, as "a bit sketchier" as a training set.

Microsoft learned of the LibGen use by April 2019, when Sam Altman and Amodei demonstrated an early GPT-3 to Bill Gates and Microsoft chief technology officer Kevin Scott, the Authors Guild says. In the summer of 2022, OpenAI deleted its LibGen files in an exercise the Guild labels "Project Clear." VP of Research Bob McGrew wrote to colleagues on June 15, 2022: "now is the right time to excise Libgen from our systems and storage."

The named plaintiffs include John Grisham, George R.R. Martin, Jonathan Franzen, Jodi Picoult, and David Baldacci. The filings also describe researcher Tarun Gogineni, hired by OpenAI in 2022, whose stated "research mission," according to the Guild, was to build a machine that would supplant human authors.