Security researchers and Anthropic have uncovered a network of around 28 fraudulent dating apps that use AI-generated personas to scam mostly male users in their mid-to-late 30s. Anthropic detected the scheme after spotting unusual activity on its Claude platform, including a prepaid account sending over 100,000 API requests daily, and disclosed the findings publicly in June before publishing a formal report. The scale and sophistication of the operation, which combines AI chatbots, real gig workers and fabricated engagement data, mark a significant escalation in AI-enabled romance fraud, and many of the implicated apps remained available on major app stores even after the disclosure.

The scam works by presenting AI personas as real potential matches on apps such as Dora, Romi and Doni, which market themselves as genuine dating services rather than AI companion apps. Only around one in four "matches" is a real person, typically a paid gig worker hired to pass video liveness checks or post pre-written replies, while the rest of the conversation is handled autonomously by AI, primarily via Claude, alongside other AI tools for reply suggestions, attractiveness scoring, moderation and avatar generation. Backend systems reportedly faked likes, profile visits and video content, and tracked which users had grown suspicious, exploiting the occasional real human contact to make the AI-driven interactions seem more believable.