The Face Database: Meta Built the World's Largest Facial Recognition System — Then 'Deleted' It

For over a decade, Facebook operated the world's largest facial recognition system, automatically identifying users in uploaded photos by comparing faces against a database of over one billion face templates. The system, which operated by default with no meaningful opt-in process, could identify individuals in photos with accuracy exceeding 97% — performance that rivaled or exceeded government facial recognition systems. In 2021, Meta announced it was shutting down the system and deleting the face templates. But the shutdown raised more questions than it answered.

What Meta Built

Facebook's facial recognition system created mathematical representations — face templates — from user photos, then used these templates to automatically tag individuals in newly uploaded images. The system processed billions of photos, learning facial features, aging patterns, and appearance variations. The resulting database represented the most comprehensive civilian facial recognition capability ever assembled, built without the informed consent of most users and without the regulatory oversight that would apply to government facial recognition programs.

The Deletion Question

Meta announced it would delete face templates for over one billion users and shut down automatic face recognition. But independent verification of data deletion is impossible — users and regulators must take Meta's word that the templates were actually destroyed. Meta retained the DeepFace algorithm and the research expertise that created the system, meaning the capability to rebuild the database persists even if the specific templates were deleted. Meta's terms of service also do not prohibit the company from reintroducing facial recognition in the future, and the company has continued developing face-related AI technology for its smart glasses and VR products.

The $650 million settlement Meta paid in Illinois for violating the state's Biometric Information Privacy Act — collecting face templates without informed consent — represented the largest biometric privacy settlement in history. But even this substantial penalty amounted to less than one percent of Meta's annual revenue, raising questions about whether financial penalties can effectively deter a company of Meta's scale from deploying biometric surveillance.

Meta's current smart glasses and VR products include camera systems capable of facial recognition, and the company's AI research continues to advance face analysis technology. While Meta states these products do not currently use facial recognition for identification, the technical capability exists and could be activated through software updates. Users and regulators should view Meta's facial recognition shutdown as a pause rather than a permanent commitment, and support legislation that requires explicit consent for biometric data collection with penalties that scale to company revenue.