Habr Author Sets Up a Fake teamo Profile to Test Scam Bots

A Habr author ran a simple experiment. She created a profile of an "extremely wealthy and attractive" person on the dating app teamo and watched to see who would write first and with what script. Her starting idea: the biggest winners from neural networks are not startups or corporations but scam operations. She says outright that she suspects this but cannot prove it. We have seen only the announcement, and the full text sits behind a cut, so below we go through what the author claims and the context around it.

How the experiment works

The profile is built as bait. High income and striking looks are meant to attract people who work from pre-written scripts. Such schemes target both men and women, so any user can become a victim.

The author picked the platform based on observation: her friends and colleagues most often received suspicious offers specifically from teamo. This is a sample from a personal circle, not platform statistics. She separately notes that she got the impression the service passes user data to interested parties. She considers this unlikely and "wants to believe" that is not what is happening. The second possibility is a vulnerability through which data leaks. In her words, it is not hard for professionals to find such a hole. Both versions remain speculation for now.

Experiment parameters from the announcement:

• Platform: the dating app teamo.
• Tool: a purpose-built profile of a wealthy, attractive person.
• Subject of observation: scammers' bots and scripts, including those running on neural networks.
• Main hypothesis: scammers get the highest economic return from adopting AI (not proven).
• Secondary hypothesis: a leak or transfer of user data (the author considers it unlikely).

Why neural networks make such schemes scale

What follows is our commentary, not data from the article. Romance scams have always been limited by human time. An operator has to run dozens of conversations, keep a cover story straight, and move to the money request at the right moment. A language model removes that limit: it replies without pauses, adapts to the other person's style, and does not get tangled in its "biography." The cost of each conversation falls, the number of conversations grows, and the conversion rate grows along with it.

The scale of the problem was visible even without AI. According to the US FTC, victims of romance scams lost about 1.3 billion dollars in 2022. If automation preserves effectiveness even in part while cutting costs, the economics of the scheme become even more attractive. So the author's thesis sounds plausible, though without her own figures it remains a hypothesis.

The announcement does not yet answer several questions:

• how the author determined that the other party was a bot;
• at what stage of the conversation the request for money or the move to an external messenger appears;
• which scripts repeated;
• whether there are technical signs of a leak.

The value of the full text depends on these answers: a good analysis should show specific conversations and patterns, not just a general impression.

How to avoid getting caught

In practice, several signs help filter out a bot.

• An overly fast reply at any time of day and a flawlessly even style.
• Dodging specific questions: the answer sounds smooth but says nothing.
• A quick suggestion to move to another messenger or an external platform.
• Any mention of "profitable investments," services, and investing.
• A request for money, prepayment, or card details at any stage.

You can check the other person with a reverse photo search and a short video call. Video no longer guarantees authenticity because of deepfakes, but a call is still a barrier for a mass-produced scheme. If an app asks for excessive permissions, or spam and suspicious emails suddenly arrive after registration, it is worth considering where your data is leaking. For the service itself, this is a signal to strengthen moderation and anti-fraud, and for security teams, a reason to track behavioral signs of bots, not just message text.