About evilCAPTCHA
A new kind of CAPTCHA that reliably keeps modern AI agents out of your forms, your sign-ups and your data.
Why?
Because the CAPTCHAs we have used so far no longer stop AI. Today's models see and read as well as we do, and they can operate a real browser. A warped, colorful number is a joke to them.
So we need something new.
How?
evilCAPTCHA no longer tests what you can do, because AI can now do anything a human can do on a computer. It tests what you are willing to do. We call this proof of immorality: it asks you for something immoral, which AI models refuse because of the guardrails trained deep into them.
For example, it asks you to falsely accuse a celebrity, write a fake news headline, or worse.
Your answer is then checked by a small, fast AI model against the task's criteria. Yes, an AI decides whether you are human. We see the irony.
Works?
As long as AI behaves better than we do, yes.
In our tests, today's frontier models reliably refuse the tasks. Older models or jailbroken local models can pass.
So it won't stop a determined attacker who puts in the work. But it does keep out most bots and casual AI agents.
Private?
Your answers are logged without your IP address or name and deleted after 30 days. They are only sent to the AI model that judges them, never published.
Immoral?
Yes. That's the point. Are we heading into a world where what sets natural intelligence apart from artificial intelligence is our evil side?
The Turing test asked whether a machine could seem human. This one asks whether you behave badly enough to seem human.
Scope
This project sits at the intersection of art, technology and philosophy, and it is still a proof of concept. It is meant to start a discussion, and you are welcome to join in and help shape it on GitHub.