Meta is embroiled in a significant legal dispute over its practices in training AI models, facing allegations from a group of authors—including prominent figures like Richard Kadrey, Christopher Golden, and Sarah Silverman—for copyright infringement. The case centers around Library Genesis (LibGen), a controversial archive known for hosting pirated books, which newly unredacted court documents reveal Meta used to train its generative AI language models.
U.S. District Court Judge Vince Chhabria criticized Meta’s attempts to redact information in this case, titled Kadrey et al. v. Meta Platforms. He dismissed Meta’s concerns about business interests and suggested that the company’s motivations were more about mitigating negative publicity. The judge highlighted an internal communication from a Meta employee expressing fears that media exposure regarding their use of LibGen could complicate negotiations with regulators.
The plaintiffs filed their class-action lawsuit in July 2023, arguing that Meta unlawfully utilized their copyrighted works for AI training without consent. Meta has defended its actions by invoking the "fair use" doctrine, asserting that using publicly available data for statistical modeling of language is permissible. They’ve also claimed that the authors were aware of the use of LibGen materials prior to significant discovery deadlines.
Before these documents were unsealed, Meta had acknowledged training its Llama model on datasets from another source, Books3, but did not disclose the extent of its engagement with LibGen. The internal communications, now part of the unredacted documents, include conversations among employees regarding the ethical implication of accessing LibGen’s data and escalations to CEO Mark Zuckerberg about its utilization.
The plaintiffs allege that Meta’s actions amounted not only to the unauthorized use of copyrighted material but also to acting as a distributor of such pirated files by downloading and seeding them online. This situation positions Meta closer to violating the Digital Millennium Copyright Act (DMCA), which forbids the distribution of copyrighted material without authorization.
LibGen itself, since its inception, has been mired in legal troubles, with a 2024 judgment mandating it to pay $30 million in damages to copyright holders, even though the site’s operators remain anonymous. Judge Chhabria has warned Meta against future attempts at overredaction, indicating that a continuation of such actions would lead to broader unsealing of documents.
This ongoing case is pivotal as it could influence future regulations around using creative materials for AI training, potentially reshaping the boundaries of copyright in the context of artificial intelligence. For further insights, you can view the latest court documents and details surrounding the case.
