Meta finds itself in an unusual legal battle after being accused of downloading thousands of adult films through torrenting networks, allegedly to train a secret AI model. The tech giant’s defense? Those downloads were probably just employees watching porn on company time.
Strike 3 Holdings and Counterlife Media, the production companies behind popular adult entertainment brands including Vixen, Tushy, Blacked, and Deeper, filed the lawsuit in July. They claim Meta illegally downloaded at least 2,369 of their movies since 2018, specifically to develop AI capabilities. The companies are seeking a whopping $359 million in damages.
The case emerged from an unexpected source. When Meta was sued by a group of authors in 2024 over alleged copyright infringement in AI training, that lawsuit revealed the company’s torrenting activities.
Strike 3 Holdings decided to check its own BitTorrent-tracking tools, which are designed to detect unauthorized downloads of their content. What they found surprised them: several corporate IP addresses traced back to Meta.
Meta Seeks Dismissal of ‘Illogical’ Porn Downloading Lawsuit
Even more intriguing, the producers allege that Meta operated a “stealth network” of 2,500 hidden IP addresses to conceal its downloading activity. According to Strike 3, this elaborate setup was meant to secretly train an unannounced adult version of Meta’s AI model, specifically for Movie Gen, the company’s generative video technology.
Meta is having none of it. The company has filed a motion to dismiss the case, armed with several counterarguments that range from technical to downright amusing.
First, there’s the timeline issue. Meta points out that 2018, when Strike 3 claims the infringement began, was years before the company even started researching large language models and generative video technology. That work didn’t begin until 2022, making it illogical that Meta would have been stockpiling adult content for AI training four years earlier.
Then there’s Meta’s terms of service, which explicitly prohibit users from generating adult content with its AI systems. The company argues it would make little sense to train models on pornographic material when the resulting AI isn’t allowed to create such content in the first place.
But the most eyebrow-raising defense centers on those download numbers. Meta argues that averaging roughly 22 downloads per year across dozens of corporate IP addresses is far too small to indicate any systematic AI training operation.
Meta Dismisses Porn Downloads as ‘Personal Use’ in AI Training Lawsuit
For context, Meta admits to using the Book3 dataset in the separate authors’ lawsuit, which contains 195,000 copyrighted books in a 37GB compilation. Compared to that scale, 2,400 videos over several years seems almost trivial.
Instead, Meta suggests a more mundane explanation: employees were simply downloading porn for their own entertainment. With tens of thousands of employees, plus contractors, visitors, and third parties accessing the internet through Meta’s network every day, the company argues it’s impossible to definitively identify who downloaded the videos or whether they were even Meta employees.
The lawsuit also mentions a Meta contractor who was allegedly instructed to download adult content at his father’s house. Meta’s response? There’s no logical reason an automation engineer would be tasked with sourcing AI training material, especially not at his dad’s place. This, the company insists, was obviously personal use.
“These claims fail not only for lack of supporting facts, but also because Plaintiffs’ theory of liability makes no sense and cannot be reconciled with the facts they do plead,” Meta stated in its filing. “The entire complaint against Meta should be dismissed with prejudice.”
Meta’s ‘Personal Use’ Defense in AI Copyright Spat Highlights Growing Legal Wave
Meta also takes a shot at Strike 3’s reputation, noting that some have labeled the company a “copyright troll” known for filing extortive lawsuits against alleged infringers.
This case is just the latest in a growing wave of copyright disputes facing AI companies. Anthropic agreed to pay authors $1.5 billion in August over alleged use of pirated books for training its language models. Apple faced similar accusations in September, with authors claiming the company used a known collection of pirated books to train its OpenELM AI system.
Whether Meta’s “it was just personal use” defense holds up in court remains to be seen. But the case highlights the increasingly complex legal questions surrounding AI training data and what companies can and cannot use to develop their systems.
