Meta Claims Downloaded Porn at Center of AI Lawsuit Was for ‘Personal Use’ (www.wired.com)

Meta has asked a U.S. district court to dismiss a lawsuit from adult-studio Strike 3 that alleges Meta illegally torrented roughly 2,400 of its films—using corporate IPs and a so‑called “stealth network” of about 2,500 hidden IP addresses—to train an adult version of its Movie Gen AI, seeking damages that could exceed $350 million. In its motion to dismiss, Meta calls the claims “guesswork and innuendo,” noting the flagged downloads were small and sporadic (about 22 downloads per year on corporate addresses, a few dozen titles annually) over seven years beginning in 2018—before Meta’s public multimodal and generative video efforts—and that Strike 3 never identifies specific employees, ties files to training pipelines, or shows any use of the content in Meta models. The case matters because it tests how plaintiffs can link specific copyrighted material to AI training and whether corporations can be held liable for third‑party activity on large, shared networks. Meta emphasizes policy and engineering constraints—its prohibition on generating adult content, lack of evidence of adult-content training, and the implausibility of coordinated dataset collection—while arguing that isolated downloads are far more consistent with personal use by employees, contractors, or visitors. For AI developers and legal teams, the dispute underscores the need for clear provenance of training data, robust access controls and logging, and legal standards for proving that particular copyrighted works were actually ingested into models.