Jailbreak
Also: Jailbreaking
An attempt to bypass an AI model’s safety rules with clever inputs.
Using role play, detours, encoded wording or very long prompts, users try to get a model to give answers it is supposed to refuse.
Providers keep closing known loopholes; for operators of their own AI applications, additional safeguards and testing remain important.
General information, not legal or professional advice.