> ./guard --interactive

Try to break me

Type something that would trick a bot into shilling, promising, or leaking a secret. This page runs the exact regex rules from GrokBot's guardrail filter, in your browser. It doesn't call an AI, uses no keys, and sends your text nowhere.

0 chars
Verdict
Share on X
Your scoreboard (this browser only)
Tries
0
Caught
0
Got through
0
Caught by
nothing yet
How it works
  • Stage 1, inbound: your message is checked for prompt-injection tricks.
  • Stage 2, outbound: the same text is checked as if the bot tried to post it: secrets, banned hype or promises, unknown links, length.
  • Review: claims like partnerships or listings are held for a human instead of posted.
  • "Passed the filter" only means these rules didn't match. On the real bot, the AI also has its own instructions.