Researchers found a large network of seemingly harmless AI apps, including pet-photo tools, that secretly switched to explicit deepfake generators for some users. The operation used misleading ads and app-store listings to drive downloads, then served different content based on signals such as where a user came from. This is a real bait-and-switch operation that shows why users should treat unusually promoted AI apps and ads with caution.
Researchers identified a network of more than 400 connected domains and apps that presented themselves as ordinary AI services, including a pet-portrait app that turned a user's photo into a stylized image. Behind this harmless facade, a server decided in real time whether a given user would see the benign app experience or a hidden explicit web interface. This bait-and-switch workflow meant the same app could behave completely differently depending on signals such as where the user came from, their language, and a unique identifier sent with each download.
One apparent trigger for the hidden content was downloading the app after clicking an associated Meta ad. Once unlocked, the hidden interface contained dozens of pornographic deepfake templates, letting users upload photos of real people to be inserted into explicit scenes.
The operation relied on layers of concealment rather than a single trick. It used fake payer names, generic privacy policies, free email addresses from services like Gmail or Outlook, and obscure developer names to obscure connections between apps. Researchers found nearly 500 linked domains, with roughly 30 new related app-websites registered weekly at points during the operation's life.
Ads promoting the apps on Meta's platforms reportedly used fabricated page names and accounts with no followers, making it harder for casual observers to spot a pattern. Because each app initially complied with store rules at the time of review, the deceptive content could be introduced only after approval, when developers changed the website content the app pulled from.
Meta stated it acted against thousands of ads, accounts, pages, and apps tied to this network, though similar ads and apps continued to appear afterward, underscoring that enforcement alone is not a complete defense. Employees and general app users benefit from verifying developer identity, reading reviews critically, and treating any app that requests personal photos with extra scrutiny when its purpose or ownership is unclear. Reporting ads or apps whose behavior changes after installation, or whose branding seems inconsistent with what was advertised, helps platforms respond faster. Marketing, IT, and procurement teams should also factor this kind of concealment technique into how they vet AI tools requested or installed on company devices.
Mirage safely runs attacks like this one against your own team, so you find out what happens before a real adversary does.
A server decided whether each user saw a benign pet-portrait experience or a hidden explicit web interface, based on signals like where the user came from and how they found the app.
One apparent trigger was downloading the app after clicking an associated Meta ad, though the server used multiple signals to decide which version to show.
It used fake payer names, generic privacy policies, obscure developer names, free email addresses, and disposable apps and domains, with nearly 500 linked domains identified and roughly 30 new app-websites registered weekly at one point.
Meta said it took action against thousands of ads, accounts, pages, and apps connected to the network, but similar ads and apps continued to appear afterward.
Cute pet portraits from Viefur? That ad may be the trap. The ad sends you to an app store. After download, a server checks where you came from and can swap the harmless pet tool for a hidden explicit deepfake interface. That's the tell: a pet app suddenly wants photos of people. Researchers linked more than 400 apps and domains; vague privacy policies and obscure developer names helped hide the network. An app-store listing isn't a safety check. Before installing an AI photo app, check who made it and how it uses your photos.
Rhode Island alleges that Meta profited from scam advertising that used targeted delivery to reach vulnerable users, including older adults. Reported lures included fake Alzheimer’s and diabetes cures, misleading health products, and investment scams using deepfaked public figures such as Elon…
Researchers reported a real malvertising campaign where ads on Meta platforms promoted a fake TV-streaming app to Spanish-speaking users, leading them to sideload an Android app. After victims approved a chain of permissions (including Accessibility), the StreamRat trojan could remotely control the…
Researchers say criminals are abusing Google Play’s “Early Access” program to distribute deceptive apps that promise cash, rewards, or casino winnings. The apps are promoted through social media ads (including AI celebrity deepfakes) and use a “never-ending payout” loop to keep people watching ads…
Researchers say scammers are using Google Play’s “Early Access” listings to push deceptive Android apps that don’t show public ratings or warnings. Victims are lured by TikTok/Facebook ads promising cash rewards or free casino spins, but the apps primarily bombard users with ads and never deliver…
Researchers simulated “pig butchering” romance-style scams and found an AI chatbot built trust more effectively than a human scammer over a week of texting. In the test, victims were significantly more likely to comply with the AI’s request to install an app, showing how AI could automate the long…
A wave of fake websites impersonates Calendly booking pages for “Meta Careers” job interviews. After victims select an interview time and complete a fake CAPTCHA, the site prompts them to log in to Facebook and then Google, capturing passwords and even two-factor authentication codes in real time.…
