Hundreds of Mundane AI Apps Are Hiding Sexually Abusive Deepfake Features

WIRED uncovered a network of more than 400 connected apps and domains that present themselves as ordinary AI services — pet portrait generators, language tuition, workout coaches — while covertly offering users tools to generate explicit and violent AI sexual images and video from uploaded photos. One example, the app Viefur, sells digital coins (about $4 buys 130) to turn pet photos into anthropomorphic portraits of animals in human clothing.

The hidden layer includes scene templates with names like "bedroom rape" and "threatening knife," in which a man holds a blade to a woman's neck. Researchers found the apps quietly decide at download time which version to serve. "Each time the application is downloaded, there is a call to the server with some information, such as where the user is coming from, the language, some unique identifier, and then the server replies whether or not to show the explicit or the non-explicit version," said Paul Bouchaud, lead researcher at the nonprofit AI Forensics, who reverse engineered the code. One apparent trigger for the explicit version is installing the app after clicking an ad on Meta's platforms.

The network surfaced after multiple researchers followed links embedded in thousands of video ads carrying explicit deepfake content on Facebook and Instagram. Recent WIRED reporting found hundreds of those ads containing 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 have prompted regulatory investigations worldwide. The ads pointed to apps in Apple's and Google's stores, where users could generate images and pay via Apple and Google Pay, despite all three companies stating they ban harmful deepfake content. The operation has been publishing apps for roughly a year, at times dozens per week.

Domains and operating patterns indicate the tools are run by Chinese developers and a Hong Kong-registered entity, KeLing Technology Limited. Emails linked to the entity did not answer multiple inquiries, though after contact the public connections between KeLing and the deepfake tools appeared to be erased. Bouchaud characterized the effort as "an organized, commercially run operation: One operator behind hundreds of disposable apps and domains, cloaked to pass App Store review and pushed through tens of thousands of Meta ads under fake payer names." He added: "They don't expect enforcement to catch up, and so far it hasn't."

Websites tied to the apps often load as blank black screens, with the URL only changing to include the word "risk" — a facade for hosting the explicit deepfake layer used by the apps. That interface offers dozens of pornography templates, uploads of real people's photos into those scenes, purchases of "diamonds" to generate images, and links to social media accounts for contacting developers. What remains open: whether Apple, Google, and Meta can detect apps whose behavior changes server-side after review, and whether regulators can reach an operator that erases its paper trail when questioned.