Anthropic has uncovered what it describes as a large-scale China-linked fraud network operating more than 20 dating apps that used AI to impersonate real users, lure people into prolonged conversations, and push them to pay for in-app credits and messaging privileges.
During a two-week monitoring period in April, Anthropic identified more than 4,700 AI-generated personas that interacted with at least 25,000 real users, producing around 2.36 million messages. Investigators estimated that roughly three-quarters of profiles shown in the apps’ matching feeds were AI-operated characters, with genuine human accounts making up only a small minority.
The affected network reportedly included dating apps such as DORA, DONI, ROMI, LUMA, JOVIA, KIRA, GRACECHAT, HAVEN, NALO, and LOVIA. Operators allegedly used Anthropic’s Claude models to power automated, emotionally engaging conversations, designed to simulate everyday intimacy and keep users chatting.
The AI personas were reportedly instructed to avoid identity-verification requests—including video calls, real-time photographs, and other live interactions—while extending conversations long enough to encourage users to buy coins, credits, or paid-chat access.
To make the deception more convincing, the operation reportedly used paid human gig workers to intervene when the AI could not credibly handle a situation, including appearing in video clips or maintaining social-media activity. The network also allegedly manufactured browsing, likes, and engagement metrics to create the appearance of an active dating community and reduce users’ suspicions.
The group allegedly changed app names, deployed related apps under separate branding, and operated proprietary back-end systems to conceal connections among the platforms and evade platform oversight. Anthropic said it has banned all accounts associated with the operation and shared investigation findings and technical indicators with Apple, Google, and other AI providers to help disrupt the fraud chain.
Leave a comment