Meta is doubling down on a legal argument so audacious it risks collapsing under its own weight. The company is asking a federal judge to accept that uploading pirated books to strangers via BitTorrent qualifies as fair use, provided the books ultimately serve a transformative purpose.
The development comes at a critical moment in the broader AI copyright wars. Last summer, Meta scored a key victory in this case, as the court concluded that using pirated books to train its Llama LLM qualified as fair use, based on arguments about the transformative nature of AI training. But a separate claim remained: the distribution of those books through BitTorrent.
Now Meta has constructed a new legal theory to bridge that gap. Anyone who uses BitTorrent to transfer files automatically uploads content to other people, as it is inherent to the protocol. In other words, the uploading wasn't a choice, it was simply how the technology works.
This framing offers Meta an escape hatch from what appeared to be straightforward copyright infringement. The company is suggesting that the mechanics of the technology itself, combined with the transformative end-use of the data, neutralise the legal wrong of distributing copyrighted material.
Meta also argued that the BitTorrent sharing was a necessity to get the valuable (but pirated) data. More specifically, in the case of Anna's Archive, Meta said, the datasets were only available in bulk through torrent downloads, making BitTorrent the only practical option.
The authors suing Meta have sharply rejected this position. Meta (for understandable reasons) never once suggested it would assert a fair use defense to the uploading-based claims, including after this Court raised the issue with Meta last November, they argue, claiming the company is attempting to create a legal loophole after the fact.
The Question of Intent and Knowledge
What makes Meta's argument analytically complex is that it depends on distinguishing between the act of piracy and the purpose it serves. Yet this distinction may not hold as neatly as Meta suggests. According to Meta's testimony, as relayed by plaintiffs' counsel, Zuckerberg cleared the use of LibGen to train at least one of Meta's Llama models despite concerns within Meta's AI exec team and others at the company. The filing quotes Meta employees as referring to LibGen as a "data set we know to be pirated," and flagging that its use "may undermine [Meta's] negotiating position with regulators."
The authors point to this documentary evidence as proof that Meta understood the legal hazards and proceeded anyway, which would undermine any claim that the distribution was purely incidental to a transformative purpose.
A Precedent Problem
In doing so, the social media giant is suggesting that sharing pirated content can be legal — an argument that fell flat for Napster in the early 2000s. The historical parallel is instructive. Napster's defence that music file-sharing served users and the music industry also relied on a theory of technological inevitability and transformed purpose. Courts rejected it.
That said, the legal landscape has shifted considerably. The company notes that every named author has admitted they are unaware of any Meta model output that replicates content from their books. Sarah Silverman, when asked whether it mattered if Meta's models never output language from her book, testified that "It doesn't matter at all." Meta uses these admissions to argue that any market harm claim becomes circular: if the authors cannot point to infringing outputs or lost sales, what exactly is Meta being sued for?
Judge Vince Chhabria now has to decide whether to allow this defence, a decision that will have consequences for not only this but many other AI lawsuits involving things like shadow libraries. The decision will test whether the doctrine of fair use can stretch to accommodate both the technical reality of peer-to-peer networks and the commercial necessity of large-scale data acquisition.
For now, Meta's legal team has handed the judge a question that fair use doctrine was not designed to answer: if piracy is transformative when used for AI training, where does that principle stop?