Meta continues to be at the center of controversies with its AI ambitions. The Facebook parent has pushed back against a copyright infringement lawsuit, denying allegations that it illegally used pornographic material to train its generative AI models. In a recent legal filing, Meta suggested that any adult content downloaded via its corporate IP addresses was likely for the ‘personal use’ of individual employees and not for model training.

The lawsuit, which was filed by Striker 3 Holdings – a company that produces and owns copyrighted adult videos, accuses Meta of illegally torrenting approximately 2,400 of its protected videos since 2018 for the purpose of training an unannounced AI model, hinting at potentially its video generator, Movie Gen. The suit also alleges that Meta concealed its activity using a “stealth network” of 2,500 “hidden IP addresses.”

Striker 3 is seeking financial damages of $350 million. 

In its motion to dismiss the lawsuit, Meta argued there is “no evidence” that it ever directed employees to illegally download adult content for AI purposes. There are no “facts to suggest that Meta has ever trained an AI model on adult images or video, much less intentionally so.”

The Mark Zuckerberg-led firm further challenged the lawsuit’s timeline, noting that the alleged illegal downloading began in 2018 — four years before its “Multimodal Models and Generative Video” research commenced.

Meta also stated that the activity on corporate IP addresses showed only about 22 downloads of adult videos per year. The company asserted that this “meager, uncoordinated activity” makes the assumption of AI training implausible. Instead, Meta argued, the more “plausible inference to be drawn… is that disparate individuals downloaded adult videos for personal use.”

The company also highlighted that its public terms of service strictly prohibit users from generating adult content through its AI models – a policy that contradicts the idea that such materials would be useful for its training data. A company spokesperson reiterated, “We don’t want this type of content, and we take deliberate steps to avoid training on this kind of material.”

Prior to this, Meta recently scored a win in a separate copyright case where a US district court ruled that its use of books to train its Llama models was protected by the fair use doctrine. Striker 3 Holdings has been given two weeks to respond to Meta’s motion to dismiss.