Jailbreak
Also known as: jailbreaking · prompt jailbreak
A jailbreak is a prompt crafted to get a model to bypass its safety or usage rules — to produce content it was configured to refuse.
- Can be roleplay, encoding tricks, or instruction manipulation.
- Targets the system prompt and safety filters.
- A core red-teaming activity.
Jailbreak vs prompt injection: a jailbreak breaks *the model's rules*; a prompt injection hijacks *the app's instructions*. Overlapping, but different targets.