400 linked apps and domains hid tools for sexual deepfakes

The app has two faces

Researchers traced the network through links in thousands of explicit deepfake ads on Facebook and Instagram. Some ads promoted AI-generated child sexual abuse material, including images of real children and a member of a European royal family. The ads reached tens of thousands of people and prompted investigations by regulators around the world.

The apps appeared in Apple and Google stores under ordinary-sounding identities, including language courses and AI fitness trainers. Their web interfaces offered dozens of pornographic templates, including scenes described as “bedroom rape” and a knife threat. Users could upload photos, generate explicit content and buy “diamonds” to pay for image generation.

The mechanism was designed to keep the ordinary storefront and the explicit service apart. Paul Boucher, a lead researcher at the nonprofit AI Forensics, said apps contacted a server when opened and sent information such as the user’s source, language and a unique identifier. The server decided whether to return the benign version or the explicit one. A click from a Meta ad may have been one trigger, though Boucher said researchers had not found a clear set of rules.

The sites also exposed how the network was assembled. Researchers found a JavaScript file listing domains and apps, app package details, encryption keys and Facebook ad trackers. Maël Le Toux, a threat researcher at Infoblox, said he found nearly 500 linked domains using the same code and referring to one another.

Boucher’s analysis of registration dates suggests the apps began appearing toward the end of 2025, then grew sharply in the spring. At one point, about 30 apps and sites for making nude images were registered each week. In September, 27 were available on Google Play; together, Boucher’s research found, they had been downloaded more than 350,000 times. Many were labelled suitable for all ages.

Enforcement came after the storefront

Apple said the apps met its rules when reviewed, then changed the site they drew content from and replaced approved material with pornography. The company said it had suspended tens of thousands of apps over the past year for similar behavior and recently updated its store rules to further restrict nude-image tools. Google did not respond to WIRED’s questions.

The network appears to have links to Chinese developers and Hong Kong-registered KeLing Technology Limited. Researchers pointed to Chinese hosting and code comments, as well as Chinese video-editing watermarks on clips used in Meta ads. KeLing’s name appeared in contact information tied to early network infrastructure and a deepfake template. A Veyra AI page, promoting a new model in late September with the slogan “More real. More natural. More uninhibited,” listed a KeLing email address.

After WIRED contacted the company, related social accounts were renamed and KeLing references disappeared from the Veyra AI site and an early network site. KeLing and addresses associated with the apps did not respond to WIRED.

Meta said it banned KeLing from advertising in 2025 after the company failed to stop its activity. After WIRED shared a list of more than 400 domains, Meta said it had found over 3,000 associated ads; two-thirds had already been removed. The company later said it had acted against 11,600 ad accounts, more than 4,700 business pages, 5,200 users and 15,000 pages linked to the domains, and barred 209 apps from advertising on its platforms.

Kathy Paul, director of the Tech Transparency Project, said the ads violated multiple Meta policies and argued that technology companies should explain where the apps’ money goes and how their operators repeatedly evade moderation and safety systems.

I think the central weakness here is not that the apps fooled reviewers once. It is that an approved app could later become a doorway to a different product, while ads, domains and developer identities were spread across platforms. Meta’s enforcement figures are substantial, but WIRED found similar ads and apps still appearing as it finished its reporting. The harder problem is keeping an approval attached to what an app actually serves, not just what reviewers saw on the day it passed.