Meta ran ads for a nonconsensual deepfake app

Researchers at the Tech Transparency Project found Meta running video ads for Kromix, an app billed as an “AI image styler” that let paying users upload photos of real people to generate sexual deepfakes, including scenarios described in the app as involving real people without their consent. One ad featured a woman resembling a prominent US politician. The 32 ads ran from a page created August 3 with no followers, targeted only at men, each drawing fewer than 10 impressions before Meta pulled the campaign after being contacted by reporters. The video ads used cover images of beaches and mountains to disguise their content from automated review.

Apple removed Kromix from the App Store, saying the app appeared to have added the prohibited content after passing initial review. Meta’s advertising policy already bans sexual imagery and the promotion of “nudify” apps; a Meta spokesperson said the company had removed 344,000 ads promoting these services from Facebook and Instagram over recent years and called the space “highly adversarial.”

What it means for you

Meta’s written policy already forbids exactly what these ads did; the gap was its mostly-automated ad-review pipeline missing content deliberately disguised behind an innocuous cover image. If you rely on a platform’s stated content policy to gauge how safe it is to build on or advertise alongside, treat enforcement rate as the real metric, not the policy text, since the two can diverge for months at real scale. This adds to a pattern of platform AI-safety failures surfacing through outside researchers rather than internal review, similar to how xAI’s Grok faced a CSAM class-action lawsuit after a similar detection gap. It’s an awkward contrast with Meta’s own product push this week, including a new Meta AI Mac app that leans on the same ad and platform infrastructure.