Judge advances authors' AI copyright lawsuit against Meta

Did tech giants intentionally scrub copyright tags off pirated books so their AI wouldn't get caught regurgitating them? A federal judge thinks there's enough smoke there to investigate the fire.

U.S. District Judge Vince Chhabria ruled on Friday that an AI copyright lawsuit against Meta can proceed. The case, brought by authors including Sarah Silverman, Richard Kadrey, and Ta-Nehisi Coates, accuses Meta of using their copyrighted books without consent to train its Llama models.

Meta had pushed hard to get the case thrown out, arguing that training on publicly available works constitutes fair use and that the authors lacked legal standing to sue in the first place.

The smoking gun behind data scrubbing

Judge Chhabria rejected Meta's push for dismissal, stating clearly that the alleged copyright infringement constitutes an obvious, concrete injury sufficient for legal standing.

Even worse for Meta, the judge allowed the authors' Digital Millennium Copyright Act (DMCA) claims to move forward. The authors argued that Meta intentionally stripped copyright management information (CMI), such as author names, titles, and copyright notices, from their books before training.

Chhabria wrote that the allegations raise a reasonable inference that Meta removed the data specifically to prevent Llama from spitting out copyright notices, which would have revealed the models were trained on protected works.

The judge did toss out one claim under the California Comprehensive Computer Data Access and Fraud Act (CDAFA). He noted the authors never alleged Meta hacked into their personal computers or servers, only that it acquired their book data.

Filings in the case have already produced embarrassing revelations, including claims that Mark Zuckerberg personally authorized the Llama team to train on copyrighted material, even as staff discussed the legal risks of using questionable datasets.

Why it matters

For authors and creators, this ruling is a massive procedural victory. AI companies have long hidden behind the vague defense of fair use, betting that plaintiffs could never prove a chatbot generated word-for-word copies of their books. By focusing on the removal of copyright management information, creators found a much sharper weapon: if you intentionally erased the copyright metadata to hide your tracks, you may be on the hook regardless of the model's eventual output.

For the AI industry, we think this should terrify data engineering teams. In the race to scrape text, stripping out metadata and headers was considered standard data hygiene. Now, a federal judge has warned that sanitizing that data looks an awful lot like concealing infringement.

Meta's aggressive bid to dismiss the lawsuit backfired by drawing attention to internal discussions where executives allegedly acknowledged the legal gray zone. While other high-profile battles like The New York Times against OpenAI play out, this ruling proves judges are willing to look past Silicon Valley's fair-use hand-waving and inspect the actual data pipeline.