Mundane AI apps hide sexual deepfake tools, WIRED finds
Original title: Hundreds of Mundane AI Apps Are Hiding Sexually Abusive Deepfake Features
Researchers tie more than 400 apps and domains to ads on Meta, with over 350,000 Play Store downloads.
WIRED reports a network of more than 400 domains and apps that advertise ordinary uses, such as the pet-portrait app Viefur, language tuition or workout coaching, then can load a web interface for explicit sexual deepfakes from uploaded photos. Templates described include scenes labeled “bedroom rape” and “threatening knife.” Researchers traced the operation from explicit deepfake video ads on Facebook and Instagram, some of which WIRED says contained AI-generated child sexual abuse imagery, including real children. Apps were listed in Apple and Google stores and took Apple and Google Pay. Paul Bouchaud of AI Forensics and Maël Le Touz of Infoblox link the sites to shared code; WIRED attributes them to Chinese developers and Hong Kong firm KeLing Technology Limited.
Unconfirmed: KeLing Technology Limited's role is WIRED's assessment from domains and operations; the firm did not respond.
- Infoblox identified almost 500 linked domains running the same code.
- About 30 related sites a week were registered at one point, Bouchaud says.
- Twenty-seven apps were live on Google Play in September and rated for all ages in many cases.
- DiffAI, created in December 2025, passed 100,000 downloads and was later removed.
- KeLing did not comment; public links to the tools appeared to be erased after contact.
- Organizations
- KeLing Technology LimitedAI ForensicsInfoblox
From the original article
8 min read at wired.com
WIRED uncovered some 400 apps and websites linked to Chinese-language developers that look like regular AI tools but had secret features that let users create violent AI-generated images and video.
By Matt Burgess · wired.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.