A WIRED investigation published on 2 November 2026 has exposed a network of more than 400 connected apps and web domains that hid tools for generating non-consensual sexual and violent deepfakes behind ordinary-looking services, including pet-photo editors, language tutors and AI workout coaches. Apple removed 19 apps after WIRED shared its findings, while Google had already pulled most of the apps before WIRED contacted it, according to reports of the investigation.
How the bait-and-switch worked
The apps passed store review by showing benign content, then contacted a remote server after installation, sending details such as the user’s language, referral source and a unique identifier. The server then decided whether to show the advertised features or load a separate web interface with deepfake generators that could be applied to photos users uploaded, with users buying credits to apply them. Researchers say one trigger appears to have been installing an app after tapping an advertisement on Meta’s Facebook or Instagram, where thousands of video ads promoted the apps. The operation was traced through a JavaScript bundle referencing hundreds of connected apps and domains, advertising trackers and encryption keys, according to reporting on the research.
Who was behind the network, according to researchers
WIRED’s review of domains points to Chinese developers and a Hong Kong-registered company, KeLing Technology Limited. That attribution comes from domain and operational analysis reported by WIRED; KeLing did not respond to WIRED. The research was led by Paul Bouchaud of the nonprofit AI Forensics, which first spotted the operation in early September 2026. In September, researchers identified 27 related apps on Google Play with more than 350,000 downloads combined, including one app, DiffAI, created in December 2025 that passed 100,000 Play Store downloads. The apps began appearing toward the end of 2025, with a spike in spring 2026.
What Apple, Google and Meta said
Apple spokesperson Adam Dema said the apps complied with App Store rules when reviewed, and that developers later changed the websites supplying their content, replacing approved material with pornographic content. Apple said it had suspended tens of thousands of apps over the past year for similar bait-and-switch tactics, and that it strengthened its rules in mid-October 2026 against non-consensual intimate imagery and hidden functionality. Meta said it banned KeLing from advertising in 2025, took action against about 11,600 ad accounts and 5,200 users linked to the network, and blocked 209 apps from advertising. Google had already removed most of the apps before WIRED made contact, according to reports of the investigation.
Why it matters
The case shows how AI-generated deceptive content can reach mainstream app stores even where policies ban it, because review checks the app as submitted, not the web content it can later load. For users, the practical warning is to be cautious about apps installed after tapping a social-media ad, especially apps that ask for photos and then direct users to a web page rather than working inside the app itself. Researchers also note that an EU ban on nudifier apps takes effect in about a month, which may change how such tools are distributed in Europe. The findings do not mean every AI photo or fitness app is unsafe, but they show why store approval alone is not a guarantee.
What is still unknown
It is not established how many users actually generated deepfake images through these apps, how many of the 400-plus apps remain available in any store, or whether the operators will reappear under new names. WIRED’s reporting attributes the network to one organisation on the basis of shared domains and infrastructure; that link has not been confirmed by the company named, which did not respond.
FAQ
How many apps were involved? More than 400 connected apps and domains, according to WIRED and AI Forensics.
What did Apple do? It removed 19 apps after WIRED shared details.
How did the apps evade review? They showed benign features during review, then loaded deepfake tools from a remote server after installation.
