The second app was never in the binary

App review inspects what a developer submits. This network kept the part that mattered somewhere else.

The apps could present their advertised features during review and to ordinary users, while making a separate server request when downloaded. According to Paul Bouchaud, the request included information such as the user's language, referral source and a unique identifier. The server then determined whether to serve the advertised experience or the hidden interface containing sexually abusive deepfake generators. Researchers had not established a complete list of switching conditions, although installing an app after following an associated Meta advertisement appeared to be one trigger. Users who reached the hidden interface could pay for generation credits through Apple Pay and Google Pay.

That design is the story. A reviewer, however thorough, cannot reject a feature that is not in the build they are given, and a user who installs the same app directly may never see anything but the advertised tool. The two populations, reviewers and targeted users, were shown different software under one listing.

Hundreds of disposable apps, one set of plumbing

The scale is what separates this from the familiar story of a few bad apps slipping through. WIRED identified a network of more than 400 connected apps and domains. Separately, Infoblox researcher Maël Le Touz mapped almost 500 linked domains, while AI Forensics researcher Paul Bouchaud tracked the apps and websites appearing across the network. The domains ran shared code and cross-referenced one another; a JavaScript bundle found by researchers contained references to hundreds of the connected apps and domains, along with encryption keys and advertising trackers. Bouchaud's analysis of registration dates found that the network began appearing toward the end of 2025. At its peak, roughly 30 new apps and associated websites were being registered each week, he told WIRED.

The downloads were not trivial. In September, researchers found 27 of the apps live on Google Play with more than 350,000 downloads between them, many rated as suitable for all ages. One, DiffAI, had been downloaded more than 100,000 times before Google removed it, shortly before WIRED contacted the company.

Paul Bouchaud, the AI Forensics researcher who reverse-engineered the cloaking method, described it to WIRED as an organized, commercially run operation: one operator behind hundreds of disposable apps and domains, cloaked to pass store review and promoted through Meta advertising. WIRED found evidence pointing to Chinese-language developers, including Chinese hosting and code containing Chinese-language comments. Separately, researchers identified connections to Hong Kong-registered KeLing Technology Limited through domain contact details, email addresses, an administration login and links to a related AI service. KeLing did not respond to WIRED's requests for comment. The investigation did not conclusively establish who controlled the entire network.

Removal followed the phone call, not the review

The enforcement record is a study in sequence. Apple removed 19 apps, including a pet-portrait app called Viefur, after WIRED contacted the company and shared the network's details. Apple's spokesperson, Adam Dema, said the apps had complied with App Store rules during their initial reviews, and that developers later changed the websites the apps loaded, replacing approved content. That is an accurate description of the mechanism, and it is also the concession: the prohibited content arrived through a channel review does not monitor after approval.

Meta told WIRED it had taken action against 11,600 advertising accounts, more than 4,700 business pages, 5,200 users and 15,000 pages linked to the domains. It also blocked 209 apps from advertising on its platforms. Google had removed most of the flagged apps before WIRED reached out, and did not respond to the magazine. Apple separately said it had suspended tens of thousands of apps over the previous year for bait-and-switch tactics. That is a company-wide enforcement figure covering such conduct generally, not a count of apps connected to this particular network.

On October 16, Apple updated its App Review Guidelines to strengthen its rules against nonconsensual intimate imagery and hidden or undocumented functionality. The changes address the two elements central to this investigation: the abusive output and the concealed features used to deliver it. They do not, by themselves, solve the problem of an approved app changing its behavior through remotely supplied content after review. WIRED reported that similar apps and advertisements were still appearing as its investigation was being completed.

Sources

  1. [1] WIRED, "400+ Apps Run Covert Deepfake Network Across Apple and Meta Stores", November 2, 2026 (paywalled; figures cross-checked against AppleInsider and AI Weekly full-text reports of the investigation)Read source
  2. [2] AppleInsider, "Hundreds of AI apps were part of a network hiding sexually abusive deepfake tools", November 2, 2026 (full-text report of the WIRED investigation, read first-hand — the more than 400 connected apps and domains, the server-side switch and its signals, the JavaScript bundle, the 27 Google Play apps and 350,000 downloads, the Dema statement, the tens-of-thousands suspension scope and the October 16 guideline update)Read source
  3. [3] AI Weekly, "Wired: 400+ Cloaked AI Apps Hid Deepfake Tools; Apple Pulls 19", November 2, 2026 (full-text report of the investigation, read first-hand — the Bouchaud statement, the roughly 30 new registrations per week at peak, and Meta's action against 11,600 ad accounts and 15,000 pages)Read source
  4. [4] Tech Times, "More Than 400 Apps and Domains Hid Abusive Deepfake Tools, Investigation Finds", November 3, 2026 (full-text report read first-hand at Gate 1 — Meta's full enforcement breakdown: 11,600 advertising accounts, more than 4,700 business pages, 5,200 users and 15,000 pages, plus 209 apps blocked from advertising)Read source
  5. [5] ALFW, "Apple's Nudify-App Ban Isn't New. Its AI Enforcement Rules Are.", October 18, 2026 (our earlier coverage of Apple's October 16 App Review Guidelines update)Read source