Meta’s Dirty Secret: Training AI with 2,396 Adult Films

Meta has been accused of illegally downloading 2,396 paid adult videos since 2018 to train its AI models, including the video‑generation model Meta Movie Gen and the large language model LLaMA.

The lawsuit was filed by two adult‑content companies, Strike 3 Holdings and Counterlife Media, in a California federal court, alleging that Meta used the pirated content both for its own training and for distribution, with more than 100,000 downloads shared.

The plaintiffs claim each infringed work could attract up to $150,000 in statutory damages, potentially totaling $359 million (about ¥25 billion). They argue that adult videos provide high‑quality, logically coherent, and low‑copyright‑enforcement material that is especially valuable for training video‑generation AI.

Meta’s defense rests on the ease of obtaining the files and the difficulty of enforcing copyright on adult content, but the companies have developed a tracking system called “VXN Scan” to monitor illegal distribution.

Meta has already admitted infringement and may settle out of court, but the reputational damage to a global tech giant could be severe.

The case follows a pattern of similar accusations against other AI leaders: OpenAI and Microsoft were sued for using copyrighted books and articles in 2023; Apple and NVIDIA faced claims of training on unlicensed YouTube videos; and Google’s use of YouTube content remains unclear.

Most of these disputes end with monetary settlements or court‑ordered compensation, reinforcing an industry norm of “train first, pay later.” However, Meta’s high‑profile lawsuit serves as a warning that legal penalties are only part of the cost; reputational harm can be far more damaging.