Meta's advertising systems ran paid promotions for Kromix, an AI tool that produces nonconsensual pornographic videos resembling female United States politicians, according to an investigation by WIRED. The ads appeared even though Meta's policies prohibit sexual material, and were removed only after WIRED asked about them.
What the Ads Promoted
According to WIRED, Kromix bills itself as an "AI image styler." A voice-over in one video advertisement promotes the tool as having "no restrictions," says "all the characters are real people," and describes it as "the AI that men actually use." In one ad, a woman closely resembling a prominent female US politician stands in front of a US flag and the flag of the president of the US above the caption "What if she moved?" She then winks, before the ad cuts to a scene of a woman with the same face performing in a pornographic video.
WIRED reported that Kromix, which was hosted by Apple's App Store prior to a request for comment, invites paid users to upload photos so the app can use them in scenarios including "bedroom rape" and "Disney love." Videos on the app also feature AI-generated pornographic videos of women wearing Spider-Man costumes. Requests for comment sent to a contact listed in the app went unanswered.
Platform Enforcement Gaps
Apple has struggled to keep similar tools, which violate its policies against nonconsensual intimate imagery, nudifying, and sexually explicit content, off the App Store. San Francisco's city attorney sent Apple a cease-and-desist letter earlier this summer demanding the removal of 13 specific apps and that Apple stop "aiding and abetting" the sale of explicit deepfake images.
The Kromix ads, according to Meta's ad library, were exclusively targeted to male users with the tagline "Unleash Next-gen AI Magic" and were removed after WIRED asked about them. Meta reported that the account tied to the campaign ran 32 ads, each running for between five and 46 hours, with most receiving 10 or fewer impressions.
| Campaign detail | Value |
|---|---|
| Total ads run | 32 |
| Ad run duration | 5–46 hours |
| Typical impressions | 10 or fewer |
| Ad page created | August 3 |
| Page followers | 0 |
Deepfakes and Women in Office
Katie Paul, director of the Tech Transparency Project, which discovered the ads, told WIRED:
"All of the ads that we've seen so far broadly on the nudify issue are exclusively sexualizing and nudifying and putting women into AI-generated porn. And this raises a lot of questions about how Meta is going to deal with this kind of content, not just broadly but when it comes to female politicians, when the company is actively running ads containing this content on their platform."
WIRED reported that politicians, overwhelmingly women, have been a target of nonconsensual deepfake images for years. In January, as Grok was used to create around 3 million sexualized images of women and girls, officials in Sweden, England, and Northern Ireland were targeted by people using the technology, often using official images as a starting point.
Benjamin Shultz, lead researcher at the American Sunlight Project, said:
"As the production and distribution of AI-generated content have democratized, formerly benign media—official portraits, video from a press conference—have become raw material for low- and no-code tools that anyone with an internet connection can use."
Automated Ad Review Under Scrutiny
Meta's public documentation says each ad on its platforms, which include Instagram and Facebook, is "reviewed" against its advertising policies before publication. "Our ad review system relies primarily on automated tools to check ads and business assets against our policies," the company's advertising standards state. The page running the advertisements was created on August 3, not long before the ads began running, and had no followers—two signals that, according to WIRED, might have indicated to Meta that the account was suspicious. The company's removal of the ads only after media inquiry, WIRED noted, marks the latest in a series of failures to stop ads for tools that produce nonconsensual intimate imagery.
Sources:
