Meta says porn downloads on its IPs were for “personal use,” not AI training

Meta has taken legal action this week, requesting a US district court to dismiss a lawsuit that accuses the tech company of illicitly downloading pornography to train its artificial intelligence systems. This legal challenge follows allegations from Strike 3 Holdings, which claims that its adult films were illegally downloaded using Meta's corporate IP addresses. The lawsuit also suggests that Meta utilized a covert network of 2,500 'hidden IP addresses' to obscure these downloads. Strike 3 Holdings is seeking damages that could surpass $350 million, asserting that Meta has engaged in stealing adult content to secretly develop an undisclosed version of its AI model, known as Movie Gen. In a motion filed on Monday, Meta countered these claims, arguing that Strike 3's accusations are based on 'guesswork and innuendo.' They also labeled Strike 3 as a 'copyright troll' known for filing extortionate lawsuits. In its defense, Meta contended that there is no substantiated evidence that the company orchestrated the downloads of approximately 2,400 adult films or was even cognizant of the illegal activities taking place. The tech firm emphasized that Strike 3 failed to provide any factual basis indicating that Meta has ever trained its AI models on adult content, let alone done so intentionally. A representative for Meta dismissed the allegations as 'bogus' when speaking with Ars. Interestingly, the downloads in question span over seven years, dating back to 2018—well before Meta's foray into research on multimodal models and generative video began. Meta argued that this timeline makes it unlikely that the downloads were intended for AI training purposes. Additionally, the company pointed out a significant flaw in the lawsuit: its own policies explicitly prohibit the generation of adult content, which contradicts the notion that such materials would be beneficial for AI training.