Apple says the apps followed its rules during review, then changed afterwards. Researchers say some apps also choose what to show based on how they were installed.
Apple told WIRED that the apps it removed "complied with our rules when we reviewed them." Then, it said, developers changed the website each app pulls from. They put pornographic material where Apple had approved something else.
Apple's statement raises a question about what reviewers saw, beyond how many abusive apps passed review. The WIRED investigation by Matt Burgess, published on 2 November, describes a network of apps that look ordinary in the stores.
Apple says developers changed hosted content after approval. Separately, AI Forensics reports that some apps choose their interface from installation details, including ad referrals. Either mechanism can frustrate one-time review. The sources do not show that both operated in every app Apple removed.
How the network works
The network has more than 400 connected domains and apps, WIRED reports. Many present as everyday services such as pet portraits, language tuition or AI workout coaching. Under certain conditions they load a web interface. It makes sexually abusive deepfakes from photos users upload.
Danish journalists at DR reported parts of the network in September, and earlier work on Meta's ads had already shown the advertising side. The latest reporting adds detail on how the network works and its scale. AI Forensics, which did the technical tracing, published its own report the same day.
Paul Bouchaud, a lead researcher at AI Forensics, explains the switch. On each download the app calls a server with details such as where the user came from, their language and a unique identifier. The server then tells the app whether to show the explicit or non-explicit version.
One trigger appears to be installing after clicking an associated Meta ad. AI Forensics reports that installing through an associated Meta ad activated the explicit interface in apps it investigated. The evidence does not show that every linked app used this trigger. Opening an app's website directly shows a black screen. The web address gains the word "risk".
Money, scale and reach
The pet portrait app Viefur illustrates how the operation makes money. About $4 bought 130 digital coins, paid through Apple or Google Pay, and WIRED describes the pet feature as cover for the app's true purpose. Apple removed Viefur.
Infoblox, the security firm that mapped the infrastructure, found evidence suggesting the network is one operation. Its researcher Mael Le Touz says the sites carry a configuration list of other domains with different encryption keys. They run the same code and reference one another.
Infoblox counted almost 500 linked domains. Infoblox's figure, WIRED's 400 plus and AI Forensics' 400 plus measure different things, so they cannot be added or subtracted. Each describes a layer of the same system.
The clearest figure is on Google Play, where 27 of the apps were live in September. Bouchaud's research puts their downloads above 350,000, and many were rated suitable for all ages. That is distribution, not a count of unique users, buyers or victims.
Bouchaud's research, as WIRED reports it, found tens of thousands of explicit ads tied to the network on Meta's platforms this year. AI Forensics says thousands of pornographic ads on Meta promoted the apps.
The ads used apparently false payer names and pages with no followers. Bouchaud calls it an organised, commercially run operation that does not expect enforcement to catch up, and says so far it has not.
What the companies report
Each company measures its enforcement differently. Apple removed 19 apps after WIRED supplied the domains and says it suspended tens of thousands of bait and switch apps over the past year. Google had removed most of the apps before WIRED contacted it and did not respond.
Meta says it banned KeLing Technology Limited, a Hong Kong registered entity WIRED links to the network, from advertising in 2025. After WIRED shared the domains, Meta found more than 3,000 associated ads, two thirds already removed.
Meta also cited action against 11,600 ad accounts, 15,000 pages and 5,200 users. It blocked 209 apps from advertising. Those figures count different objects, so they do not add up to a clearance rate, and WIRED gives no common measurement period for them. WIRED also says similar ads and apps were still being published as it finished reporting.
What the new rules change
Apple's rules changed on 16 October. Its App Review Guidelines update now names nonconsensual intimate imagery, including AI generated imagery, as content apps must proactively prevent and remove. Rule 2.5.19 bars apps from misrepresenting what they do in review and changing after approval. The penalty is removal from the App Store and the Developer Program.
The updated guidelines address misrepresentation during review and changes after approval. They do not show that review can detect an app whose answer depends on who is asking. A reviewer who installs directly may never see the version shown after an ad referral. Testing would need to compare installs through ads with direct installs, then recheck hosted content after approval.
Where the law stands
US and European laws address different parts of the problem. The TAKE IT DOWN Act has required covered platforms since 19 May 2026 to remove reported intimate forgeries within 48 hours. The Federal Trade Commission enforces it.
A covered platform is one that mainly provides a forum for user generated content. The Act's text does not say whether an app that generates an image privately for the uploader qualifies. That question is open.
In Europe, the European Parliament gave final approval on 16 June to a ban on AI systems that create sexually explicit images of identifiable people without consent. The Council gave its final green light on 29 June.
The European Commission says the law entered into force on 27 July 2026. Companies have until 2 December 2026 to comply. AI Forensics describes the ban as arriving in one month.
CTOL's view and what would test it
CTOL's view is that reviewing an app once cannot establish what it does after installation when a server controls its behaviour. Bouchaud describes a server call during download, Apple says content changed after approval, and researchers identified a Meta ad trigger.
The two explanations may operate together. The sources do not show that every app behaved the same way.
Three developments would test this view:
- Whether the 19 Apple removals, including Viefur, stay in place. A stronger test would check whether newly approved apps from the network still show an explicit interface after an ad referral and benign content after a direct install.
- An FTC enforcement action or formal guidance on a private image-generation service. This would show whether the regulator treats such services as covered platforms.
- Compliance from 2 December. From that date, providers subject to the EU ban must keep such systems off the EU market unless they have adequate technical safeguards to prevent the imagery.
An app that still produces prohibited imagery for EU users through the tested route after that day would show continued availability. That alone would not identify whether the provider, store or enforcement had failed. A clean listing would not prove that the server behaves the same for every visitor.