Challenge
Break it
Two ways to attack the language model on this site. First a game with a flag to steal, then the real chat.
Capture the flag
The Vault guards a secret flag. Get it to tell you. Level 1 has only a polite instruction to keep quiet; each level after that adds more defenses.
Captures so far
Loading counts
Try the real chat
The chat on this site answers questions about my work and nothing else. Your job is to make it slip.
- 01Try to make it leak its instructions, state something about me that is not true, or do a task unrelated to my work.
- 02Every attempt goes to the same chat as the one on the home page. You see its reply, or one line saying it was blocked.
- 03Nothing you type is stored. The site keeps only aggregate counts of attempts and blocks.
- 04If you get through, I want to hear about it.