Wild WordAI Experience
jailbreak
“A prompt or trick that gets a model to ignore its safety rules and do what it was trained to refuse.”
First Seen
2022
Context
Borrowed from phone unlocking, AI jailbreaks emerged with ChatGPT's launch. Techniques like 'DAN' (Do Anything Now) use roleplay, hypotheticals, or token manipulation to bypass safety guardrails. The cat-and-mouse dynamic between jailbreakers and safety teams reveals fundamental tensions in alignment.
Citation
A wild word — documented from usage in the field, not coined by Xenolexica.
Ethics & Intention