A decade-old data sourcing deal has forced an AI company to purge its training sets. Clarifai has deleted 3 million OkCupid profile photos it obtained in 2014 and confirmed to the FTC it also scrapped any machine learning models trained on that data. The deletion follows a settlement between the FTC and OkCupid's parent company, Match Group, over a violation of the dating site's own privacy policy.

According to Reuters, court documents show Clarifai directly requested the data from OkCupid executives, who provided it. The company used the images to develop facial recognition models for classifying age, gender, and race. For data engineers, this case highlights the long-term technical debt of training data provenance. Models built on improperly sourced data can become liabilities, requiring full retraining from scratch years later—a costly architectural consequence.

The FTC's investigation began in 2019 after reports of the data use. Clarifai founder Matthew Zeiler had argued for public trust in tech companies' use of such data. The settlement permanently bars OkCupid from misrepresenting its data privacy controls, a provision some observers note merely restates existing FTC rules. For practitioners, the episode serves as a stark reminder: system design decisions around data acquisition must account for regulatory and ethical constraints, not just immediate model performance. The data pipeline's origin matters as much as its output.

Source: Engadget