Put a safety check in front of chat
Safety check in front of chat that blocks jailbreaks and explains why.
Guarded chat
Type a prompt below. The guard checks it first, then the chatbot answers when it is safe.
Your prompt goes to Muse Spark 1.3 for the safety check and the reply. This demo keeps history in this browser only.
- Safety checks are heuristic. They can miss new tricks or flag safe text.
Safety log
Each prompt and reply check lands here with its category and reason.
- History and safety logs stay in this browser only.
Guard settings
Strictness
Code caution
When code caution is on, the guard also screens prompts for risky code patterns.
- Code review is basic pattern screening, not a full security audit.
Teams that need full control can self-host open protection models instead of this check.
Idea from a Meta developer video.