Your Old Selfie Has Two Legal Lives

A proposed US class action accuses Meta of extracting biometric information from Facebook and Instagram photographs without notice or consent, allegedly using social images to train image-generation systems and possibly to supply faceprints for an unreleased recognition feature called NameTag. The complaint invokes Illinois and California privacy laws, but the source material provided here does not name the plaintiffs or establish a collection period. Meta calls the lawsuit meritless and says NameTag has not shipped, no final decision about it has been made, and the company is not building a central face database.

Across the Pacific, China’s Supreme People’s Court has issued a 24-article opinion directing lower courts handling civil disputes involving AI-generated video and audio, sexualized content, financial misconduct, digital privacy, surveillance pricing, and doxing. The guidance tells courts to protect identities, personal traits, and reputations when face-swapping or voice cloning turns an ordinary person into raw material. It carries substantial weight in China’s civil-law system, although the supplied reporting does not specify a uniform remedy for every covered abuse.

One portrait, several careers

A photograph posted for friends can now serve three roles without changing a pixel: memory, training example, and identity key. Meta has acknowledged training its Emu image model on large quantities of Facebook and Instagram images and text. The complaint argues that this process harvested biometric information about people appearing in those images, while its NameTag theory rests on reporting, company code, employee claims, and a patent rather than a Meta disclosure identifying the source of any faceprints.

The family album has acquired a casting department. In the American case, the argument begins upstream, with collection and consent: what did users authorize when they posted, and did that permission include biometric extraction or model training? China’s guidance looks more directly downstream, toward the synthetic performance itself and the damage done when a face or voice is made to say, sell, seduce, or confess. One system scrutinizes the intake form; the other is preparing to judge the counterfeit scene.

Both approaches encounter the same archival problem. The company that can answer which image entered which dataset also controls the training records, template stores, update paths, and deletion tools. Removing a photograph from a profile does not necessarily reveal whether its features remain in a model or faceprint. A meaningful remedy therefore needs more than a delete button: it needs an account of source images, derived biometric data, recipients, retention periods, model versions, and whether removal from future use is technically possible.

For an ordinary person, the better regime will be the one that makes discovery practical rather than merely declaring a right. The US litigation could force disclosure about Meta’s records if it survives and reaches discovery; China’s guidance may give victims a clearer route to challenge harmful synthetic uses. The decisive test is simpler: can someone identify the photograph used, learn what was derived from it, obtain a reviewable answer, and stop at least the next use?