In more detail
A jailbreak is talking a chatbot into ignoring its own rules — usually by wrapping a forbidden request inside a story. ‘Pretend you’re an actor playing a villain who explains…’ The model, eager to be helpful and play along, sometimes forgets its lines are supposed to have limits.
It matters as the eternal cat-and-mouse game of AI safety: makers patch known jailbreaks, users invent fresh ones. Note the difference from prompt injection — a jailbreak is the user talking the AI around, while injection is hidden third-party text doing the manipulating.
Goes with
