Meta has added a new wrinkle to its legal defense in a copyright lawsuit brought by a group of authors in 2023: uploading pirated books to other BitTorrent users during a download isn't infringement, the company argues — it's fair use.
The argument appeared in a supplemental interrogatory response filed in California federal court. Meta's position is that BitTorrent seeding, the process by which a downloading client simultaneously redistributes data to other users, was an unavoidable feature of the protocol rather than a deliberate act of infringement. Because bulk datasets from shadow libraries like Anna's Archive were only practically obtainable via torrent, Meta argues the seeding was inseparable from the same transformative use the court already approved.
The court ruled last summer in Meta's favor on the core question: using the books to train its Llama language models constitutes fair use. The BitTorrent distribution claim — who got what from whom — was the one live issue remaining from the lawsuit originally filed in 2023 by Sarah Silverman, Richard Kadrey, Christopher Golden, and others.
Meta's supplemental filing also leans on deposition testimony from the plaintiffs themselves. Silverman testified that whether Meta's models ever reproduce language from her books "doesn't matter at all." Meta is now using that admission to argue there's no demonstrated market harm — a key factor in any fair use analysis.
The authors' attorneys filed a letter with Judge Vince Chhabria calling the late Friday submission a procedural end-run: a new defense introduced after discovery had closed. They say Meta had been aware of the uploading claims since November 2024 and never raised fair use in connection with them, even when the court raised the issue directly. Meta responded the next day, pointing to a December 2025 joint case management statement where it says the defense was explicitly flagged — and noting that plaintiffs' counsel addressed it at a subsequent hearing. Whether Judge Chhabria treats timeliness as a threshold question may determine whether the fair use argument ever gets a substantive hearing.
The broader implications are considerable. A ruling that peer-to-peer data acquisition can shelter under a technical-necessity fair use theory would reshape how AI companies approach large-scale training data sourcing. Meta also cited U.S. global AI leadership as a public interest rationale — a framing sweeping enough to attract scrutiny of its own.