You are chatting with someone who takes an interest in your life and replies day and night. But is there a person on the other side? That is exactly what our Spajk highlighted when sharing Anthropic's findings. A fluent conversation no longer verifies the other person's identity.
What Anthropic uncovered on dating sites
According to Anthropic's September 2026 report , a China-based studio operated a network of more than twenty dating apps. The services presented themselves as dating platforms with real people, but included AI-controlled profiles. Over two weeks in April, Anthropic recorded more than 4,700 such profiles, conversations with at least 25,000 people and approximately 2.36 million messages generated by Claude. These are findings from one specific monitored network, not a count of fake profiles across all dating sites.
Hired people were part of the offering too. There was approximately one real person for every three AI profiles, available to take a video call or verify a social media account. The operator used this to dispel doubts and, according to Anthropic, even tracked who began to suspect they were chatting with a bot.
Why would someone create thousands of fake profiles?
In the network described, the main source of money was the communication itself. Users paid for additional messages using in-app coins. The longer an attractive profile kept them talking, the more they could spend. AI made it possible to conduct thousands of conversations simultaneously, around the clock and at much lower cost than human staff. Profile photos, fake expressions of interest and avoiding video calls were all part of a system designed to make the relationship seem credible. Source: Anthropic.
This differs from a scam involving a direct request to transfer money. In both cases, however, a person invests trust in a relationship whose terms the other party conceals. Anthropic's report does not say that everyone contacted lost their savings, or that the perpetrators were the same as those in Czech cases.
When paid chat turns into a romance scam
Instead of charging for messages, a scammer may build a personal relationship and only then make an urgent request. One time they cannot afford to visit you; another time they need to pay for supposed medical treatment, customs duties on a parcel or help for a child. Offers of a “great investment” also appear. The story changes according to what works on the particular person. The Police of the Czech Republic explainthat, besides money, the goal may be to obtain sensitive information or use the victim's bank account for transfers.
AI makes it possible to write naturally, quickly and at scale. But not every fake profile is AI-controlled: there may be a person, a group or a combination of people and automation behind it. A photo, voice message or short video call alone does not prove identity. In the network described, some verification calls were handled by hired people.
A Czech case: a fake celebrity and millions of crowns
The risk is not limited to foreign dating sites. In 2025, Prague police described the case of a seventy-year-old man who believed he was chatting on social media with actress Sandra Bullock. The person on the other end promised him marriage and a life together in the United States. The man sent money through gift cards; in total, he lost more than CZK 6 million and even sold his home. The case reported on the basis of police information does not prove that AI was used. It does, however, show how easily a false identity and the promise of love can lead to serious harm.
Czech police classify romance scams as a long-term form of manipulation. Scammers pose as famous people, doctors or soldiers, and often move the conversation from a dating site to private messages. Published Czech cases therefore demonstrate the same type of risk for people in the Czech Republic, not the presence of the particular app network uncovered by Anthropic.
How to recognise warning signs and what to do
Be alert if the other person quickly talks about love or a future together, but keeps postponing an in-person meeting. Other warnings include pressure to pay for more messages, keeping secrets from family, moving to another platform or requests for money, gift cards, cryptocurrency or account access. No single sign is foolproof; the overall pattern of behaviour matters.
Before sending money, pause and discuss the situation with someone you trust. Verify a famous person's identity outside the chat itself, through their publicly listed channels. Do not share login details or allow someone else's money to pass through your account. If you have already paid, contact your bank immediately, save the conversation and payment records, and contact the police. There is no reason to feel ashamed because someone abused your trust.
Online today, a reply may come from a real person, an automated system, a hired worker or a scammer. So the question is not just “Do they write like a person?” but also “Do I know who is talking to me, why they are keeping me in the conversation and what they want from me?”
Back to all articles