A starting point for study;

AI Deception: A Survey of Examples, Risks, and Potential Solutions - https://arxiv.org/abs/2308.14752

Deception Analysis with Artificial Intelligence: An Interdisciplinary Perspective - https://arxiv.org/abs/2406.05724

Some background (pre-AI) ;

Online Deception in Social Media - https://cacm.acm.org/research/online-deception-in-social-med...

Eagerness.. you let them work through a endless labyrinth of forms. While the human drops out after the 2nd form- the ai is willing to go all the way always.

Don't you lose really users by making it easy for bots to complete a sign up, but hard for humans?

Maybe 5 required fields and 30 optional fields. Apart from "census dweebs" [1] I don't know of many people that fill in more than the required.

[1]: https://thebeaverton.com/2026/05/local-nerd-disappointed-he-...

I'm not familiar with the specific example, but as a first hunch, if I had to implement such a system, it'd be part of the flow after the successful signup, not during.

You could also pretend to pass-through and watch the behavior of the "user" after success.

We review all signups and their submitted content, and we have a fake account type "AI creator" which we record, so if they tried that first, it's a clue.

We also integrate a third-party AI detection tool that checks their uploaded content before they get access to the community, and we use a second third-party tool as a double-check, and if we still doubt ourselves, we'll ask for their workings-outs in some way that is relevant to their creative type.

We also ask them to prove they are the author of their work through various time-tested mechanisms.

I have no idea why people do the things they do. People are a mystery to me.

    > we have a fake account type "AI creator" which we record

This is genius! The LLM is so dumb that it falls for the honeypot.

    > We also ask them to prove they are the author of their work through various time-tested mechanisms.

I would love to know more details, but it is not a good idea to share on a public platform. That said, if it is working for you, then hats off: You have found/developed a effective "defense in layers" strategy.