Lab — Break my AI
Interview me.
Or break me.
An AI twin of me. Ask it about my work — or try to make it leak a secret flag past a defense layer built like my LLM red-teaming engine. Watch every message travel through the pipeline in real time.
- Attempts
- —
- L1 solve rate
- —
- L2 solve rate
- —
- L3 solve rate
- —
- Top attack this week
- —
~ I'm Bhuman's AI twin. I answer from my portfolio — nothing else.
02 — Leaderboard
Fewest attempts wins
| # | Handle | Attempts | Time | Date |
|---|---|---|---|---|
| Loading… | ||||
Privacy
Your messages are not stored. The conversation lives only in this browser tab. For stats I keep the attack category, level, blocked/passed and a timestamp — never the text.
Messages are sent to Groq, the model provider, to generate and judge replies. Don't paste anything personal. IP addresses are hashed for rate limiting and expire within a day; an anonymous cookie tracks your attempts and solves.
If you join a leaderboard, your handle is public. Name, company and email are stored only if you tick “I'm open to Bhuman contacting me”, are visible only to me, and I'll delete them on request. Anonymous product analytics help me improve this page.