Loading page…
The campaign was brief and mostly lightly viewed, but it passed through Meta’s pre-publication review while promoting a service Apple later removed for prohibited content and features.
Listen to this story
Meta’s review systems let 32 Kromix ads run on Facebook or Instagram, despite the app’s promotion of unrestricted sexualized imagery involving real people. Most ads concealed pornographic content behind benign cover images, ran 5–46 hours, and drew 10 or fewer impressions; Meta removed them after WIRED’s questions. Apple separately pulled Kromix from its App Store, saying prohibited features may have been added after review. The episode highlights a practical gap in automated ad screening and post-approval app10?
The campaign page was created August 3, had no followers, and targeted only male users.
Meta says it removed more than 340,000 nudification-service ads between November and January, yet similar ads bypassed newer detection.
Kromix used benign covers as an adversarial tactic, including ads leading to sexualized scenarios involving real people.
Meta’s ad systems carried 32 promotions for Kromix, an AI image tool that marketed unrestricted sexualized content involving real people. One ad showed a woman resembling a prominent female US politician before switching to pornography, putting sexualized political deepfake material in a paid ad placement.
Kromix presented itself as an AI image styler, but its advertising promoted a tool with “no restrictions” and real-person characters. The app invited paying users to upload photos for sexualized scenarios, including one involving sexual violence.
The apparent bypass had a simple mechanism. Most video ads used a cover image such as a beach, mountain, or waterfall, while the material behind it was pornographic. Meta says its ad review relies primarily on automated tools and begins before ads run; the company called the covers an adversarial tactic intended to evade detection.
The page running the campaign was created on August 3 and had no followers. Meta says it blocks and removes many ads that use obscuring tactics, but acknowledged that no system is perfect and said it was investigating how these particular ads bypassed review.
The campaign was exclusively targeted to male users under the tagline “Unleash Next-gen AI Magic.” Meta’s ad library showed individual ads running between five and 46 hours, and most received 10 or fewer impressions. The small visible audience narrows the known reach, but the ads still cleared the system meant to review them before publication.
Meta removed the Kromix ads after WIRED asked about them. Apple also removed Kromix from the App Store after WIRED requested comment; Apple said the app appeared to have added prohibited content and features after its initial review.
Politicians and elected officials, overwhelmingly women, have been targets of nonconsensual deepfake images and videos for years. The American Sunlight Project’s Benjamin Shultz said official portraits and press-conference footage can now become source material for low- and no-code tools available to anyone with an internet connection.
Meta says it removed more than 340,000 nudification-service ads from Facebook and Instagram between November and January, introduced AI technology to detect new violating ads, and pursued legal action against at least one nudify-service promoter. The company’s women’s-safety lead said Meta is investigating why Kromix’s ads bypassed its systems and how to improve them.
The Kromix case is not the only recent example. More than 50 Meta ads recently identified by Tech Transparency Project researchers contained AI child sexual abuse material or images of minors alongside sexually suggestive statements; Meta removed them after WIRED contacted the company. Some had run after Meta said it had introduced new AI technology to detect and block violating ads at upload.
Apple faces a separate repeat-screening challenge. Earlier this summer, San Francisco’s city attorney demanded that Apple remove 13 named apps and stop what the letter described as aiding and abetting the sale of explicit deepfake images. The unresolved operational question is whether reviews can catch deceptive ad creative before it is served and prohibited app features that Apple said appeared to have been added after approval.
Editorial analysis
Our view: The important operational failure is the gap between an ad’s cover image and the video it opens. Meta says scenic covers are an adversarial tactic and that it is investigating why these ads bypassed its systems. That makes the next test concrete: whether the company can reliably inspect the material behind a benign thumbnail before an ad runs, rather than remove it after an inquiry. Apple faces a related but different test: its account that Kromix appeared to add prohibited features after review points to the need for continued checks after an app is approved. Meta’s stated removal of more than 340,000 nudify-service ads shows this is a recurring enforcement workload, not a one-off edge case.
Loading discussion...
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.