Meta argues BitTorrent “uploads” are just technical noise — asks court to treat them as fair use

New legal tack in high‑profile AI copyright fight

Meta has told a federal court that the incidental uploads produced by BitTorrent are an unavoidable technical feature — and therefore part of the same fair‑use activity that a judge already found lawful when the company trained its LLaMA model on pirated books. It has been reported that Meta and other tech firms downloaded copyrighted works from pirate repositories such as Anna's Archive (安娜档案库) via BitTorrent to assemble training datasets, and the company is now advancing a novel defense: if downloading to build a transformative AI model is fair use, then the simultaneous, automatic sharing that BitTorrent forces on users should be treated the same.

Background: the narrow issue left in the case

The dispute began as a 2023 class action by prominent writers — including Richard Kadrey, Sarah Silverman and Christopher Golden — who allege large‑scale copyright infringement. Last summer Meta won a partial victory when the court accepted arguments that using entire books in training could be a transformative fair use. But one issue remained: authors say Meta’s BitTorrent downloads also uploaded copies to strangers, which they call direct, large‑scale infringement. BitTorrent is a peer‑to‑peer protocol: downloading typically means you also upload pieces to other peers. Reportedly, Meta’s lawyers now argue that those uploads are not a separate, intentional act of infringement but a necessary and inseparable part of a lawful, transformative use.

Procedure, pushback and geopolitics

Authors’ lawyers responded angrily, saying the upload defense was raised late and was an improper attempt to dodge discovery deadlines under Federal Rule of Civil Procedure 26(e). Meta countered that the argument is not new and pointed to prior case management filings and testimony from the writers themselves, which Meta says undercut claims of market harm. The company also framed part of its defense in geopolitical terms, arguing that its AI investments have helped secure U.S. leadership in the global AI race — a national‑security angle that adds weight to what might otherwise look like a narrow technical dispute.

What’s at stake

Judge Vince Chhabria must now decide whether “technical necessity” converts peer‑to‑peer sharing into covered conduct under fair‑use doctrine. The ruling will matter far beyond one company. Will courts allow the mechanics of file‑transfer protocols to blunt copyright protections for authors? And what precedent will this set for the many other AI developers who have reportedly used pirated or scraping‑only datasets to train models? The answer could reshape the legal map for AI training data and for how copyright law adapts to distributed technologies.