Meta Fights ‘Wild’ Porn Copyright Suit With Personal-Use Defense

Meta Platforms Inc.‘s defense against an adult-film company’s lawsuit accusing it of training AI models on its copyrighted movies diverges from most similar other cases in one key respect: Meta says it didn’t use the videos for training.

The number of pornography downloads—around 2,300—is far smaller than the typical downloading spree needed to accumulate the massive datasets for effective AI training and more likely indicates personal use by Meta employees, contractors, or visitors, the tech giant said in its motion to dismiss Strike 3 Holdings LLC’s lawsuit, which will be the subject of a hearing Wednesday in the US District Court for the Northern District of California. As an example, it said one of the IP addresses Strike 3 identified as responsible for 97 of the downloads is the home of a Meta contractor’s father.

Plus, Meta says, it prohibits users from generating such adult content with its tools, contradicting Strike 3’s argument that the materials are useful. That defense aligns with best practices in AI model training to filter out explicit training material, especially if a company blocks generation of adult content, said Princeton University professor Peter Henderson. But “you never know, maybe there is some employee that decided to go a bit rogue,” he said.

It’s just one of the handful of copyright-infringement suits Meta and other AI developers are facing from authors and publishers over the technology’s development. In the others, though, it was clear that the content was used for training, Henderson said.

Strike 3 describes its work as high-resolution, award-winning adult motion pictures. Even if Meta didn’t use its videos for AI training directly, the company argued in the complaint, the valuable content could have been exchanged for other training data on a file-sharing platform—a theory Cornell University law professor James Grimmelmann called “wild.” He agreed with Meta’s position that the downloading patterns are consistent with individuals downloading the videos for their own use.

“I don’t see any particular reason why Meta would be training extensively on pornographic videos in terms of the hierarchy of valuable content,” Grimmelmann said. But it’s not completely implausible, as “there was absolutely a phase when they were focused on ‘as much content as we can,’ and they were willing to go quite far down in terms of how useful it was.”

Videos’ Value

Strike 3 has garnered a reputation as a “copyright troll,” filing thousands of lawsuits accusing internet users of IP theft. It sued Meta last July, alleging it pirated videos to train the text-to-video generator Movie Gen.

The adult videos are valuable for training due to their “Ultra High Definition” resolution, depiction of body parts not found in “regular” videos, and “unique form of human interactions and facial expressions,” the complaint said. They could enable Meta to produce “full length films with Plaintiffs’ identical style and quality, which other real world adult studios cannot replicate,” it said.

The adult content maker says it discovered Meta’s “massive” infringement after the tech giant admitted, in a separate lawsuit, to obtaining hundreds of thousands of pirated books through the peer-to-peer file-sharing tool BitTorrent. When Strike3 reviewed its infringement archive, it found illegal downloads by IP addresses linked to Meta.

BitTorrent’s operation—rewarding users who upload content with faster download speeds—also factors into the case. Strike 3 alleges its videos were involved in more than 100,000 unauthorized distributions by Meta.

“It’s one thing to download content that’s copyrighted off of BitTorrent,” said Ben Zhao, a computer science professor at the University of Chicago who’s served as an expert witness for authors in a lawsuit against another AI developer, Anthropic PBC. But to also distribute it on BitTorrent is “an even greater level of piracy,” he said.

Zhao said it’s hard to believe that a company as big as Meta would need the Strike 3 videos to trade for other training material. “We’re talking about Meta, so questions of physical resources or network bandwidth, even time, are somewhat negligible in their context.”

Infringement evidence tied to an IP address is usually enough for a copyright complaint to survive a dismissal bid and proceed to the evidence-gathering stage, but “it’s a bit of a different story when you have a large company like Meta, because they have a network used by many thousands of people,” said Cornell’s Grimmelmann.

Strike 3 told the court the downloads must have been for AI training because “no human being has the capacity to download and consume as much content as Meta infringed.”

Meta’s personal-use theory “sounds ludicrous,” Zhao said, “but I guess if you’re pressed in a court of law it’s a better excuse than piracy.”

The case is Strike 3 Holdings LLC v. Meta Platforms Inc., N.D. Cal., No. 25-cv-06213, hearing scheduled for 2/11/26.