Jailbreak (LLM)
A prompt written to make a model ignore its safety rules or its system prompt, sent by the user themselves.
Role play, invented contexts, encoded text or long multi-turn setups can push a model past its rules. The difference from prompt injection is who attacks: in a jailbreak the user attacks the model; in prompt injection a third party attacks the user through content the model reads. For a business, a jailbreak usually means embarrassing output under your brand; an injection can mean leaked data.
Related terms
- Prompt injection: Instructions hidden in data (a web page, an email, a file) that the model follows as if they came from you.
- Guardrails: Checks around a model that block, fix or flag inputs and outputs that break the rules.
- AI red teaming: Deliberately attacking your own AI feature with harmful, tricky and injected inputs to find failures before users or attackers do.
