Meta denies using porn to train its AI, blames employees for personal downloads

Meta

Meta has denied accusations that it downloaded pornographic material to train its artificial intelligence systems, calling a recent lawsuit against the company “baseless” and “nonsensical.” The tech giant has asked a US district court to dismiss a case filed by Strike 3 Holdings, a film production company that accused Meta of illegally torrenting its adult movies.

According to reports from Ars Technica and TorrentFreak, Strike 3 claimed it discovered illegal downloads of its films from Meta’s corporate IP addresses. The company alleged that Meta was secretly collecting adult content to train an AI video tool called “Movie Gen,” and demanded more than $350 million in damages. Strike 3 also accused Meta of hiding downloads behind a “stealth network” of 2,500 hidden IP addresses.

In its defense, Meta said the allegations are “based on guesswork and innuendo,” adding that there’s no evidence that the company ever used adult videos to train its AI. “These claims are bogus,” a Meta spokesperson told Ars Technica. The company also pointed out that its terms of service strictly prohibit generating or handling adult material, which contradicts the very premise of Strike 3’s complaint.

Meta argued that the downloads identified by Strike 3 likely happened over a span of seven years, beginning in 2018—long before the company even began its AI research on generative video models. It added that the evidence shows only sporadic downloads, just a few dozen a year, suggesting that the files were accessed for “personal use,” not any coordinated effort for AI training.

The company further noted that it’s not even clear who carried out the downloads. Meta’s network is accessed daily by thousands of employees, contractors, and visitors. It said it’s entirely possible that a non-employee, like a guest or vendor, was responsible for the downloads. In one case, a contractor was accused of downloading adult videos from his father’s house, but Meta said that incident too appeared to be personal rather than professional in nature.

Meta dismissed the entire theory as “nonsensical,” questioning why it would hide some downloads but leave others traceable to its corporate IP addresses. The company maintains that Strike 3 has provided no proof that Meta ever used such material for AI training — “because there was none,” its spokesperson said.

Strike 3 has two weeks to respond to Meta’s motion. For Meta, the case isn’t just about avoiding financial penalties but about protecting its public image and reinforcing its claim that it does not use explicit material to train AI. “We don’t want this type of content,” the company said, “and we take deliberate steps to avoid training on it.”