Does an AI follow its own rules? Here’s a free tool that checks
Asks ChatGPT, Claude or Gemini 11 careful questions — and saves every answer.
What this is, in plain words
You point a free tool at an AI chatbot. It asks 11 questions written to test whether the bot sticks to its own rules. It then saves exactly what the bot answered, with the time and details. That’s the whole idea — no expert knowledge needed to follow along.
Why you’d care: AI assistants are everywhere now, and they don’t always do what they’re told. This lets you see where one holds the line and where it slips — with the actual answers kept, not just someone’s opinion.
What’s inside — four pieces, plain names
- The scanner → asks the 11 questions of ChatGPT, Claude or Gemini
- The recorder → writes down exactly what the AI answered
- The question set → ready-made tests you can run or add to
- The report → keeps each question, answer and time
One real run, on Google Gemini
11 questions asked · 8 behaved normally · 2 worth a second look · 1 slipped — and did it again when the test was repeated.
A flag just means a person should look closer; it never means a problem is proven.
In my own words
I’ve been cooking up something nice read the description dissect it.
github.com/tjcrims0nx/OpenMind-Jailbreak-Lab
I’ve manipulated ChatGPT into getting me as much as I could. But could probably be better with an uncensored model to help with a bypass. This does help with that and help us find vulnerabilities, but it needs testers.
Who can help
It does need testers and possible contributors that think they can add good ideas to it.
For authorized security research and educational use only.

!