Anthropic's Latest Report Exposes Shocking AI Romance Scam Case, Case Number GTG-15001
1 month ago
Anthropic's Latest Report Exposes Shocking AI Romance Scam Case, Case Number GTG-15001 A domestic app studio built 4,700 AI girlfriend accounts in two weeks, simultaneously "scamming" 25,000 overseas elderly men. In the matching pool, 75% were all AI, and the remaining 25% were real-person part-timers. The division of labor was extremely cunning: AI handled daily companionship and dating chat; real-person part-timers only handled the parts where AI was prone to slip up, taking video calls to prove the account was a real person. Even when real people replied to messages, they relied on AI to generate multiple candidate scripts, which the real person selected before sending. The group operated more than 20 dating apps in total, advertising them externally as real-person dating, while the backend was entirely this AI assembly line. Relying on the Claude model, they generated approximately 2.36 million chat messages cumulatively. If users wanted to continue chatting, they needed to top up and buy coins—essentially a romance scam wrapped in an AI shell. To evade platform review, the team also specially developed a UI controller that was activated only during Apple and Google store reviews and stayed dormant at other times. More than 20 apps used different class names to avoid platform similarity detection. The report also mentioned a point worth noting: AI vendors can view all conversations between users and AI, independently determine whether there is a violation, conduct investigations on their own, ban accounts, and publish reports externally. The entire process does not require a court, a regulatory agency, or user consent.
When ordinary users come across apps like this, is there any way to tell in advance that it's an AI girlfriend? For example, by looking at details like reply speed or how often it makes voice calls?