FTC widens AI enforcement
- U.S. regulators signaled they will enforce AI rules beyond competition, targeting harms like deepfakes and voice cloning. - Clarifai said it deleted three million OkCupid photos and trained facial models after FTC scrutiny and a settlement. - Regulators are now able to force deletion of training data, turning data provenance into an operational vulnerability (reuters.com).
The Federal Trade Commission is pushing its artificial intelligence enforcement beyond monopoly issues and into how companies get, use, and delete training data. (reuters.com) That shift came into view after Clarifai told Reuters it deleted nearly three million OkCupid user photos and facial-recognition models trained on them after Federal Trade Commission scrutiny and a settlement with OkCupid and Match Group. (reuters.com) The Federal Trade Commission sued OkCupid and Match Group Americas on March 30, 2026, alleging OkCupid shared photos, location data, and other information with an unrelated third party despite privacy-policy promises to users. The agency said the dataset included nearly three million user photos. (ftc.gov) OkCupid and Match Group settled without admitting or denying the allegations, and a stipulated order bars them from misrepresenting their privacy practices. The order says the third party was given access to photos and other data without formal limits on how the information could be used. (ftc.gov) Artificial intelligence systems learn from examples, so a photo set can become part of the model itself after training, much like ingredients baked into bread. When regulators force a company to delete both the source files and the models built from them, they are attacking the training pipeline, not just the original database. (reuters.com) (ftc.gov) The Federal Trade Commission has used that remedy before. In its 2021 Everalbum case, the agency required the company to delete photos and videos, face-recognition data created without consent, and models or algorithms developed in whole or in part from that material. (ftc.gov) The agency has also been building a separate record on voice cloning, where software copies a person’s speech patterns closely enough to sound like them on a phone call. In April 2024, the Federal Trade Commission said scammers were using cloned voices in family-emergency and business-impersonation schemes, and it ran a challenge for tools to detect or watermark fake audio. (consumer.ftc.gov) (ftc.gov) The result is a wider enforcement map for artificial intelligence companies: not only whether a model works, but whether the underlying photos, voices, and other personal data were obtained on terms a regulator can defend in court. Reuters reported that Clarifai said it complied and deleted the data and models. (reuters.com) For companies building artificial intelligence systems, the risk now sits upstream. A dataset that looked like an asset can become a liability if the Federal Trade Commission decides the consent trail does not hold up. (reuters.com)
