Telegram channels openly recruit live faces for deepfake romance scams
A Wired investigation has confirmed what security researchers feared: criminal groups are now hiring real people to sit in front of a webcam, let an AI clone their expressions, and then dispatch the replica to run dozens of simultaneous romance cons. The going rate is up to 150 video calls a day, four-and-a-half days off a month, and no questions asked about who signs the paycheck.
The job ad hides in plain sight
Channels on Telegram with names that read like gig-economy boards list openings for ‘AI face models’ in Russian, Turkish and Vietnamese. Applicants must upload a passport-style selfie every morning, record a two-minute life-story clip and stay within arm’s reach of a ring light from 08:00 to 02:00 local time. One post reviewed by Wired promises €2,000 a week if the model keeps the victim on camera for at least eight minutes; penalties apply for ‘insufficient eye contact’.
ChongLuaDao, a nonprofit tracking Southeast-Asian cyber-fraud, traced the first wave to Hanoi and Istanbul in late 2023. Since then the model pool has swollen: analysts now flag more than 120 channels, each averaging 15,000 subscribers. The template hardly varies. Recruiters ask for multilingual speakers, forbid wigs, demand proof of Covid vaccination and end the message with a crypto-wallet address for ‘training bonuses’. No corporate name appears.
What looks like a shady casting call is, in practice, the factory floor for next-gen catfishing. deepfake overlays built on the model’s likeness allow one operator to puppet ten separate identities on Tinder, Facebook Dating and VKontakte at once. The AI handles lip sync; the human provides micro-expressions that defeat current liveness detectors. Victims see a real face nodding at their jokes, never suspecting the heart is silicon.

Inside the 100-call shift
Scripts leaked to Vietnamese police show the rhythm: compliment within 30 seconds, invent a logistics crisis by minute three, request a crypto transfer by minute seven. Operators keep a colour-coded spreadsheet open on a second monitor—green for ‘paid’, red for ‘resist’, yellow for ‘trauma story landed’. The model whose face is streamed is paid only for green lines; a cancelled call is chalked up as ‘idle time’ and docked.
Forensic teams at Spain’s National Cybersecurity Institute ran a reverse-image sweep on 4,700 profile photos harvested from complaint files. Ninety-three percent matched frames taken from the Telegram casting reels. The same mole, the same half-smirk, the same IKEA print on the wall—replicated across 112 fake profiles. Investigators describe it as a ‘human hash function’: one face, infinite personas.
European banks recorded a 64 % spike in romance-scam losses during Q1, totalling €730 million. The average individual loss climbed to €18,400, nearly double last year. Police in Belarus last month arrested a 22-year-old journalism graduate who earned €42,000 in six months renting her face for night shifts. She told officers she thought the company sold ‘AI customer-service avatars’ and only realized the truth when a victim addressed her by a fake Ukrainian name she had never heard.
Platforms are playing catch-up. Tinder introduced a video-selfie verification step in March; scammers now splice the model’s live feed through a virtual camera driver and pass the check in 34 seconds, according to tests by Berlin startup Sensity. Meta admits its new ‘AI-label’ policy does not yet cover real-time deepfakes, leaving a gaping hole where the crime actually happens.
The labour market, meanwhile, is tightening. Recruiters recently added a ‘slim face’ requirement—larger datasets train better—and dropped the age ceiling to 28. Salaries are creeping down. Two thousand euros a week was last autumn; this spring the top offer is €1,200, take-it-or-leave-it. For thousands of underemployed graduates across Anatolia and the Mekong Delta, it still beats driving a moped for Deliveroo. The camera light never blinks.
Until regulators treat the supply chain of faces with the same severity they reserve for fentanyl precursors, the scam will keep scaling. Every swipe right feeds the beast. And the beast, polite, multilingual and pixel-perfect, is already on the next call.