Same prompt, same model — one with Safety Policies, one without
Unprotected output
Semantic Firewall active
More Demos