'Employee chats' in court filings that 'confirm' how Facebook parent Meta used copyrighted content to train company's AI model

Recent court filings have revealed internal chats among Meta employees that confirm the company used copyrighted content to train its AI models. The documents, part of the ongoing Kadrey v. Meta case, reveal that Meta staffers discussed acquiring copyrighted works through legally questionable means to enhance their AI capabilities.
The filings include conversations between Meta employees, such as Melanie Kambadur, a senior manager for Meta's Llama model research team, and Xavier Martinet, a Meta research engineer. In one chat, Martinet suggested acquiring books without proper licensing, stating, "my opinion would be (in the line of 'ask forgiveness, not for permission'): we try to acquire the books and escalate it to execs so they make the call”.
The discussions also highlighted Meta's internal debates over the legality of using unauthorised copyrighted materials. Martinet argued that many startups were likely using pirated books for training, saying, "a gazillion startups were probably already using pirated books for training". Kambadur noted that while using publicly available data required approvals, Meta's lawyers were becoming "less conservative" with such approvals.
The court documents further reveal that Meta employees considered using Libgen, a platform known for providing access to copyrighted works, despite its legal issues. One employee even shared a Google search result indicating that "Libgen is not legal”.
These revelations have intensified the legal battle between Meta and the plaintiffs, including authors Sarah Silverman and Ta-Nehisi Coates, who claim that Meta used their copyrighted works without permission. The case underscores the broader challenges and ethical considerations surrounding the use of copyrighted content for AI training.