Clarifai removes 3 million OkCupid images used for facial recognition AI training, report says

The AI company Clarifai has deleted around 3 million images originally sourced from OkCupid after it was revealed that the data had been used to train a facial recognition system, according to a Reuters report. The company also confirmed that it removed any AI models trained on the same dataset.

The data sharing came under scrutiny following an investigation by the U.S. Federal Trade Commission (FTC). According to findings referenced in court documents and reported by Reuters, Clarifai had requested access to OkCupid user data in 2014. OkCupid, a dating platform whose executives had previously invested in Clarifai, reportedly provided user-uploaded photos along with demographic and location information.

This data transfer allegedly conflicted with OkCupid’s privacy policy at the time, which should have prevented such use.

In an email cited in court records and reviewed by Reuters, Clarifai founder and CEO Matthew Zeiler wrote to OkCupid co-founder Maxwell Krohn: “We’re collecting data now and just realised that OKCupid must have a HUGE amount of awesome data for this.”

Although the data-sharing activity reportedly occurred over a decade ago, the issue resurfaced in 2019 when a New York Times article reported that Clarifai had used OkCupid images to develop AI tools capable of estimating attributes such as age, gender, and race from facial images. This led the FTC to open an investigation.

The FTC’s inquiry, along with claims involving Match Group (OkCupid’s parent company), culminated in a settlement last month. While neither OkCupid nor Match Group admitted wrongdoing, Clarifai’s decision to delete the dataset has been seen as an acknowledgement that the data was in use.

Regulators also alleged that Match Group and OkCupid failed to disclose the data-sharing practices and attempted to hinder parts of the investigation. The settlement prevents OkCupid and Match Group from misrepresenting or facilitating misrepresentation regarding how user data is collected or shared. They emphasised that, while it could not impose financial penalties for a first-time violation in this case, the companies are now permanently restricted from engaging in similar practices.