Meta mocked for "Bob Dylan defense" in AI copyright battle

If you have been following the avalanche of AI copyright lawsuits, you are probably familiar with the industry's default shield: training an AI model on internet data is protected by fair use.

Meta is finding out that shield has a massive hole in it. The authors suing the tech giant are not just arguing about weights and outputs; they caught Meta red-handed using BitTorrent to download massive caches of pirated books.

Caught seeding on the open web

On Monday, authors filed for summary judgment on direct copyright infringement in a California federal court. Their argument is simple: GenAI or not, downloading pirated material for corporate gain has been illegal for decades.

According to the authors, Meta originally tried downloading pirated books individually, but the process crippled internal networks and ran far too slowly. Desperate to catch up in the AI race, Meta switched to BitTorrent to grab terabytes of pirated datasets at once, allegedly routing the downloads through Amazon Web Services to avoid detection.

That decision created a fatal legal trap. The BitTorrent protocol fundamentally relies on peer-to-peer uploading. Even if Meta throttled seeding after finishing a file, BitTorrent clients continuously upload partial files during the download phase, a behavior known as leeching. By doing so, Meta allegedly distributed pirated book chunks back to other internet pirates.

The authors mocked Meta's posture as the Bob Dylan defense, quoting his song Sweetheart Like You: Steal a little and they throw you in jail; steal a lot and they make you king. In a twist of irony, the pirated files Meta grabbed included books written by at least 10 Supreme Court justices.

Even Meta engineers knew it was wrong

Meta is scrambling to keep the torrenting evidence out of the spotlight. The company argued against allowing the plaintiffs to introduce new expert reports on leeching, claiming it has not had an adequate chance to investigate the theory.

The plaintiffs also pushed to depose CEO Mark Zuckerberg again, arguing newly unsealed emails show the torrenting scheme was escalated all the way to executive leadership. Meta flatly refused an additional deposition for Zuckerberg, offering two other senior executives instead.

Meta did agree to make six employees available ahead of the May 1 hearing. Among them is software engineer Nikolay Bashlykov, who candidly admitted in an internal message at the time that torrenting from a corporate laptop does not feel right.

Judge Vince Chhabria admitted in a prior hearing that leeching was an unfamiliar term to him. He will decide on May 1 whether it would be unfair to Meta to rule on summary judgment at this stage.

Why it matters

By focusing on BitTorrent leeching, the authors found a brilliant backdoor around the complicated fair use debate. Judges may struggle to understand neural network representations, but they understand Napster and peer-to-peer piracy.

There is zero legal precedent in American copyright law protecting the unauthorized upload and distribution of copyrighted works over torrent networks as fair use. The authors do not even need to win the argument over model weights; proving Meta distributed portions of their books to third parties over BitTorrent could be enough to secure a direct infringement ruling.

We think this exposed a glaring operational vulnerability inside frontier AI labs. The internal message acknowledging that downloading torrents on corporate laptops felt wrong is devastating evidence of willful disregard. If Judge Chhabria sides with the plaintiffs on the leeching theory, it could undercut the fair use defenses of any AI developer that relied on peer-to-peer piracy networks to feed its foundation models.