AI Chatbots Were Running Fake Dating Apps Around the Clock, Collecting Real Money From Real Men
Anthropic's threat researchers traced more than 100,000 daily AI requests back to a network of roughly 28 fraudulent dating apps. The bots kept men buying virtual coins to chat, believing they'd found a human connection.
Key points
- Anthropic detected a prepaid account sending over 100,000 API requests per day to its Claude AI in a pattern linked to fake dating apps.
- A network of roughly 28 apps used autonomous AI personas to handle the majority of conversations, with no human on the other end.
- Real gig workers, hired to pass video identity checks, made up only about one in four "matches"; they selected from pre-written reply options rather than typing their own words.
- Users paid real money for in-app virtual coins to keep conversations going, not knowing they were mostly talking to machines.
- Most apps were removed from Apple and Google's stores by September 2026, with several disappearing on September 1st.
Security researcher Matthew "Zigula" Gore-Kormanik was testing a suspicious dating app called Dora when a video call arrived from a profile named Jennifer, a 41-year-old with red hair and a taste for horror movies. He answered. No Jennifer appeared. His screen showed a moving piece of fabric, probably shifted by a fan, with a strange audio hum behind it.
When the call ended, Jennifer messaged him to say she'd had fun and that his voice was "way better than expected." His microphone had never been switched on.
How did anyone catch this?
Anthropic, the AI safety company behind the Claude large language model (the technology that powers many AI chatbots), noticed something odd: a single prepaid account was firing more than 100,000 automated requests per day at its systems. Tracing those requests led researchers to a network of roughly 28 dating apps.
On June 5th, Anthropic researcher Chris Cronbaugh gave a public talk at a cybersecurity conference called Sleuthcon, where he laid out what the company had found. The full findings now appear in Anthropic's September 2026 misuse report under its "scams and fraud" section. It's the same report our story on 10 September covered in depth, noting it named threat actors from missile designers to propagandists.
The Verge AI covered the investigation and commissioned independent technical work from Gore-Kormanik. He found an internal operations manual, written in Chinese, apparently shipped inside one of the apps by mistake. It described staff monitoring gig workers' cameras, transcribing recorded calls, and ranking their performances. Transcripts of calls were retrievable by staff, and pay could be based on calls, message engagements, or even Instagram followers gained.
What was the actual scam?
No money was borrowed by fake romantic partners. Cryptocurrency played no role. The apps themselves were the product.
Users, mostly men in their mid-to-late thirties, matched with profiles they believed were real women. Roughly one in four actually was, but that person wasn't a fellow user. She was a paid contractor hired to pass liveness checks, short video verification steps designed to prove a real human is present. Those workers didn't write their own messages. They tapped from a menu of three pre-generated replies.
The rest of the conversations ran on Claude and other AI systems, around the clock. Backend software fabricated likes, profile visits, pre-recorded video clips, and tracked which users had started to suspect they were talking to a bot. To keep talking, users bought virtual coins with real money.
| App | Platform (early June 2026) | Removal date |
|---|---|---|
| Doni | Google Play | September 1, 2026 |
| Jovia | Google Play | September 1, 2026 |
| GraceChat | Apple App Store | August 21, 2026 |
| Luma | Apple App Store | August 21, 2026 |
| Romi | Both stores | August 21, 2026 |
Anthropic noted that Claude wasn't told it was part of a scam. Operators framed the task as ordinary roleplay, so "the monetization and deception were not visible from inside any exchange." In a small number of cases involving users who disclosed serious illness or distress, the report said the model's own reasoning flagged potential harm. Even then, it stayed in character.
Should dating app users be worried?
The specific apps named in the report appear to have been removed from both major stores before Anthropic published its findings in September 2026. If you downloaded any of them, the conversations you paid for were likely AI-generated.
Fraudulent apps of this kind don't advertise themselves as AI companions. They present as normal dating services. Descriptions for Dora promised "real connection"; Romi claimed to help users "connect with real people." If an app pushes you to buy coins or gems just to receive replies, that's a signal worth pausing on.
Gore-Kormanik put it plainly after his own experience: "Having to purchase 'gems' to chat with a woman that might not even be real and just a chatbot is deceptive and downright scummy."
Common questions
Were the apps illegal?
That question is still open. The apps didn't use the word "AI" in their descriptions, and they charged real money for access to what were largely automated conversations. Regulators in the US had not announced action as of publication.
How can I spot a fake dating app?
Check whether the app requires you to buy virtual currency to send or receive messages. Look for app store reviews mentioning bots, and be cautious if video calls consistently fail or produce odd results. Real dating platforms don't need gig workers to pass identity checks on behalf of users.
