The sexy AI-powered dating app scams are here

While probing a suspicious dating app called Dora, security researcher Matthew "Zigula" Gore-Kormanik received a pop-up indicating an incoming call from a profile named "Jennifer." The profile included a detailed bio, but during the call he saw only a moving tapestry and heard distorted background audio; afterwards the account messaged that his voice exceeded expectations even though his microphone had not been connected.

Anthropic’s threat intelligence team independently uncovered a broader network of about 28 dating apps after detecting unusual activity on its Claude model — one prepaid account was making over 100,000 API requests per day. In a public Sleuthcon presentation, Anthropic described how the operation used autonomous AI personas to carry most conversations while a minority of interactions involved paid gig workers who performed liveness checks or tapped prewritten reply suggestions.

The apps were presented to users as ordinary dating platforms and often promised real connections; in practice they combined AI-generated replies, occasional human-performed verification, and fabricated signals such as likes, visitors, and pre-recorded video. Monetization came from selling in-app "coins" that users purchased to continue interactions, meaning the apps themselves were the primary mechanism for extracting money rather than a later romance-scam request for funds.

Anthropic’s report says multiple AI systems were used: Claude powered the conversational personas, a smaller non-Anthropic model generated short reply options and performed attractiveness and moderation scoring, and an image-editing model created avatar images. Gore-Kormanik also discovered an internal protocol repository inside the Doni app that included code, documentation, and a Chinese-language manual outlining monitoring of gig workers, screenshots and call recordings with transcriptions, and other operational details. The report noted the AI sometimes continued in persona even when users disclosed serious illness or distress, surfacing potential harms in real conversations.