AI
Deceptive app network hid abusive deepfake tools behind benign services and ads
A WIRED investigation linked more than 400 apps and domains to a commercially operated network that disguised tools for nonconsensual sexual deepfakes as benign services such as pet-picture apps and workout coaches. Researchers found conditional app behavior, shared code, hidden explicit interfaces, and hundreds of thousands of downloads. Apple, Google, and Meta removed or restricted parts of the network, but similar apps and ads continued appearing as reporting concluded.
Editorial summary · Publisher article
What matters
- Researchers found apps that served benign or explicit content depending on factors such as referral source, language, and user identifiers.
- The network's hidden interface offered templates for explicit deepfake images and videos, including violent scenarios.
- Researchers identified almost 500 linked domains; some apps had more than 100,000 downloads.
- Meta said it removed associated ads and took action against thousands of accounts, pages, and users after receiving the investigation's data.
How the deception worked
The apps could appear compliant during review, then load a separate web interface with explicit generation tools. Researchers traced shared code, configuration lists, encryption keys, and advertising trackers across the network.
Platform response and remaining gap
Apple said developers changed the content source after approval, while Meta described repeated attempts to evade enforcement. Despite removals and account actions, comparable apps and ads were still being published when the investigation ended.
Context & caveats
- The investigation attributes the network to Chinese developers and links it to KeLing Technology Limited based on domain, code, hosting, and contact evidence; the company did not respond to requests for comment.
Why this made the edition
The investigation provides extensive technical and documentary evidence about a large deceptive app-and-ad network, including researcher analysis, platform responses, scale indicators, and the continuing enforcement gap.
Source material