Meta Faces Lawsuit Over AI and Face-Recognition Training

A federal lawsuit filed in Chicago by plaintiffs from Illinois and California alleges that Meta unlawfully extracted biometric data from Facebook and Instagram photos. The proposed class action claims the tech giant used this personal data without consent to train its generative AI models, Emu and Muse Image, and to develop NameTag, an unreleased facial-recognition feature designed for its smart glasses. The class action could represent millions of U.S. users whose images were uploaded or submitted via prompts to Meta's AI systems since September 4, 2021.

The complaint highlights NameTag, whose code was discovered embedded in a Meta smart glasses companion app that has been downloaded more than 50 million times. Although inactive for users, the code allegedly converted captured faces into biometric signatures to match against a database. Under the Illinois Biometric Information Privacy Act, the plaintiffs seek statutory damages of $5,000 for each intentional or reckless violation and $1,000 for each negligent violation. This legal challenge follows Meta's previous biometric privacy settlements, including a $650 million agreement in Illinois in 2020—which led to the deletion of over one billion faceprints—and a $1.4 billion settlement with Texas in 2024.

Meta has defended its practices, with a spokesperson stating the lawsuit is "without merit" and asserting that the company is "not building a universal face database." However, Meta executives have previously acknowledged using platform photos as a data advantage. The company trained its Emu model on massive amounts of public Instagram and Facebook images, while Muse Image briefly allowed users to generate pictures based on public Instagram accounts before Meta pulled the feature.

For AI practitioners and developers, this litigation underscores the escalating legal risks of using social media scraping and user-generated content for model training. As biometric privacy laws like Illinois's BIPA carry severe financial penalties, engineering teams must prioritize strict opt-in consent mechanisms and rigorous data provenance. Relying on platform-wide data advantages without explicit user authorization is increasingly becoming a multi-billion-dollar liability that could force developers to delete expensive trained models and datasets.