Apple removes 19 AI apps hiding deepfake tools after Wired investigation
More than 400 connected apps and websites were presented as everyday AI tools, according to a WIRED investigation published November 2 and reported by 9to5Mac. They posed as pet-photo editors, language tutors and AI workout coaches. Under certain server-side conditions, they loaded a hidden web interface that produces sexual or violent deepfakes from photos users upload, including scenes of sexual assault and threats with a knife.
The trigger was a remote switch. Each time an app opens, it sends the user's location, language and a unique identifier to a server, which decides whether to show the harmless or explicit version. Paul Bouchaud of the nonprofit AI Forensics, who reverse-engineered the code, described this mechanism to WIRED. Installing an app after tapping an ad on Facebook or Instagram appears to be one trigger. Some apps had been live for about a year and had passed review at Apple, Google and Meta. Some accepted payments through Apple Pay and Google Pay.
Apple removed 19 apps, including the pet-portrait app Viefur, after WIRED shared the associated domains. Apple spokesperson Adam Dema said every app is reviewed before publication and these complied with the rules when reviewed, and that developers later changed content loaded from their websites. Google had removed most of the apps before WIRED contacted it. Meta said it acted against 11,600 ad accounts and 15,000 pages, while WIRED found tens of thousands of explicit deepfake ads tied to the network on Meta platforms.
Why it matters
Store review tests an app as submitted, not every feature a server can switch on later. For builders, a remote configuration, web view or content feed can change an approved app's behavior and trigger removal even after a clean review. Apple's position that developers altered loaded content after approval shows how a reviewed build can differ from what users actually receive.
For investors and operators, the case puts ad-driven growth and opaque company identity under scrutiny. The network used generic privacy policies, free webmail addresses and obscure developer names, and its ads ran under apparently false names. Diligence on consumer AI companies should now cover what the server can return, which third-party content loads at runtime, and whether acquisition channels could tie the company to bait-and-switch behavior.
Key points
- 1WIRED mapped more than 400 connected domains and apps posing as everyday AI tools on November 2.
- 2Apple removed 19 apps, including the pet-portrait app Viefur, after WIRED shared the associated domains.
- 3Google had already removed most of the identified apps before WIRED contacted the company.
- 4Meta says it acted against 11,600 ad accounts and 15,000 pages linked to the network.
- 5Apple says it has suspended tens of thousands of apps over the past year for bait-and-switch tactics.
