Unsealed court filings in Authors Guild v. OpenAI have produced what may be the most damaging internal communications yet surfaced in any AI copyright case. The documents, filed last Thursday as part of a motion for partial summary judgment, do not merely suggest OpenAI used copyrighted material — they show executives and employees discussing their knowledge of the illegality, their indifference to the consequences for authors, and their active efforts to destroy evidence. The case is no longer about whether AI training on copyrighted works constitutes fair use in the abstract. It is now about documented intent. The most structurally significant revelation concerns the chain of knowledge. Sam Altman and Dario Amodei disclosed the use of LibGen — a pirate library hosting millions of copyrighted books — to Bill Gates and Microsoft CTO Kevin Scott as early as April 2019, during a presentation of an early GPT-3 build. Microsoft's subsequent investment and partnership with OpenAI proceeded with this knowledge in hand. Amodei's internal characterization of LibGen as 'a bit sketchier' and researcher Sam McCandlish's worry about 'optics' — specifically imagining the Hacker News headline 'OpenAI uses copyrighted data from sketchy Russian website' — frame the concern as reputational, not legal or ethical. The internal communications about authors' livelihoods are equally stark. OpenAI Policy Director Jack Clark wrote in May 2020 that their work 'will make people unemployed' and predicted they would 'likely ignore' artists' concerns 'and release anyway.' Tarun Gogineni, hired in 2022 to improve GPT's writing quality, described his 'research mission' as having GPT models finish George R.R. Martin's unfinished book series, musing that 'even if GRRM dies early, GPT-5 will autocomplete his series.' Gogineni viewed authors' complaints about stolen datasets as 'acceptable economic disruption' and predicted 'the death of the reader' as 'machines created slop for more machines.' The evidence destruction angle introduces potential spoliation issues. OpenAI's internal 'Project Clear' involved deleting LibGen files in summer 2022. Slack messages show OpenAI Corporate Designee Paino asking 'how concerned are we about mentions of libgen?' across internal systems, followed by VP of Research Bob McGrew directing the excision of LibGen 'from our systems and storage' because 'OpenAI is in the news.' If the court finds this constitutes spoliation of evidence, the procedural consequences could be severe — adverse inference instructions would effectively tell the jury to assume the deleted material was damaging. The plaintiffs' roster — including John Grisham, Jonathan Franzen, George R.R. Martin, Jodi Picoult, and the Authors Guild's 18,000 members — ensures this case carries cultural weight alongside its legal significance. Lead counsel Justin Nelson from Susman Godfrey is steering the class action through multidistrict litigation in Manhattan, with additional briefing expected over the coming months and a hearing anticipated in early 2027. The legal question of whether AI training on copyrighted works constitutes fair use remains unsettled, but these filings shift the terrain dramatically. Fair use analysis weighs the purpose and character of the use, the nature of the copyrighted work, the amount used, and the market effect. Internal statements from OpenAI's own employees — acknowledging market substitution, anticipating author displacement, and describing the training data as stolen — hand the plaintiffs evidence on virtually every prong. Microsoft's exposure is compounded by its documented early awareness and subsequent multi-billion-dollar investment. The broader structural pattern here extends beyond one lawsuit. If courts find that AI companies cannot claim good-faith fair use when their own executives documented knowledge of infringement and market harm, the entire industry's training data strategy faces repricing. Every major foundation model company has similar exposure. The question is whether the legal system moves fast enough to impose consequences before the extraction is complete and the authors' market position is irrecoverable.
