Millions of OkCupid photos fed into AI models prior to FTC intervention

Nearly three million user photos from OkCupid were used to train facial-recognition AI models years before regulators intervened, highlighting a prolonged gap between data misuse and enforcement in the U.S. privacy landscape.

The images, along with location and demographic data, were shared in 2014 with AI company Clarifai without user knowledge or consent. The transfer occurred without a formal agreement or defined limitations on how the data could be used, despite OkCupid’s privacy policy at the time explicitly stating that personal data would not be shared with unaffiliated third parties.

The dataset was subsequently used to develop facial-recognition systems capable of identifying individuals and analyzing attributes such as age, race, and gender. The scale and sensitivity of the data—sourced from a dating platform—have raised concerns about how personal images were repurposed for AI development outside their original context.

The issue remained unaddressed for years. The Federal Trade Commission opened an investigation in 2019, but the case was only resolved in 2026. As part of the settlement, Clarifai confirmed it had deleted both the three million images and any AI models trained on them, formally certifying the action to regulators.

OkCupid’s parent company, Match Group, is now prohibited from misrepresenting its data practices for 20 years. However, the settlement does not include financial penalties, reflecting limitations in the FTC’s authority under existing law.

The extended timeline between the initial data transfer and regulatory action underscores ongoing challenges in policing how consumer data is used in AI training. It also raises unresolved questions about how long the models were in operation and whether similar datasets have been used elsewhere in the AI ecosystem.