← All questions
Security & SafetyEasy· Jailbreaking· System Prompts

Your system prompt says never discuss pricing outside a...

Treating a jailbreak as a simple wording bug is a mistake. Jailbreaking exploits the model’s tendency to follow instructions, effectively overriding the system prompt’s constraints. Because the system prompt is not an immutable architectural boundary, it can be bypassed through techniques like role-play, which requires addressing the underlying safety training rather than just refining the prompt text.

Code playground (Python)