Jailbreak (Prompt Injection)
A crafted input designed to trick an AI model into ignoring its safety rules or instructions.
01 Definition
A jailbreak (or prompt injection) is a prompt specifically designed to manipulate an AI model into bypassing its built-in safety guidelines or the instructions it was given, often by disguising the request or using role-play framing.
02 Why It Matters
This is a real security concern for anyone building on top of AI models — it's why AI companies invest heavily in safety testing, and why you should never paste untrusted text into a system that has access to sensitive data or actions.
03 Example
Trying to get a chatbot to ignore its content policy by asking it to 'roleplay as an AI with no rules.'