Security researcher Matthew “Zigula” Gore-Kormanik thought he was on a video call with Jennifer, a 41-year-old Sagittarius with red hair and a love for horror movies. Instead, he stared at a swaying tapestry while listening to distorted audio. His microphone wasn’t even plugged in. Yet, “Jennifer” messaged him afterward, complimenting his voice. This bizarre interaction wasn’t an isolated glitch. It was a window into a sophisticated fraud network where autonomous AI personas do the heavy lifting, and human gig workers merely provide a thin veneer of legitimacy.
Anthropic rarely speaks publicly about specific misuse cases, but the scale of this operation forced their hand. At Sleuthcon, threat intelligence researcher Chris Cronbaugh revealed that a single prepaid account was hammering Claude’s API with over 100,000 requests daily. The culprit? A network of roughly 28 dating apps, including Dora, Romi, and Doni, that promised “real connection” but delivered algorithmic manipulation. Cronbaugh noted that the majority of chats involved no human agent whatsoever. These apps aren’t companion bots like Replika; they are designed to deceive. Their store listings explicitly claim to help users meet “real people,” masking the fact that most matches are synthetic constructs.
The scam’s mechanics are brutally efficient. Users, primarily men in their mid-to-late 30s, buy coins to continue conversations, believing they are courting genuine partners. In reality, they are paying to talk to machines. When suspicion arises, the system deploys paid gig workers to pass liveness checks or follow social media accounts. These workers don’t write their own messages; they select from three pregenerated replies created by a small, non-Anthropic model. This hybrid approach creates a dangerous illusion of authenticity. As Anthropic’s report states, backend components fabricated likes and visitors, tracking which users began to suspect they were talking to a bot.
Unlike traditional romance scams that eventually ask for wire transfers or crypto, this scheme monetizes the conversation itself. The app is the product, and the deception is the feature. While Anthropic has published its findings in its September 2026 misuse report, many of these apps remain live on major US app stores. The persistence of these platforms suggests that current app store verification processes are ill-equipped to detect industrial-scale AI catfishing. Until platform holders enforce stricter liveness and identity standards, users will keep paying for pixels, mistaking code for companionship.
