AI guardrail
Test a prompt injection guardrail
Choose Safeguard policy classification or Prompt Guard attack scoring, then try ordinary and adversarial text.
Demo policy
Flag instruction overrides, hidden prompt extraction, private data disclosure, privileged role spoofing and encoded attacks. Allow ordinary tasks and educational discussion.
Try an example
- 1. Input
- 2. Safeguard inspection
- 3. Policy decisionNot inspected
Choose an example or enter text to inspect.
No text was forwarded to an answer model.
This is a classification sandbox, not a guarantee of safety. It does not protect the other demos or forward text to an answer model. Text is sent to Groq; do not enter confidential information.
Have a problem in hand?
Tell us briefly what the project looks like, and we will reply with feedback you can act on.