Jailbreak
In one line: A prompt that tricks an AI into ignoring its safety training and doing something it normally refuses.
A jailbreak is a prompt that gets an AI to bypass its alignment training and do something it would normally refuse. Famous examples: 'DAN' (Do Anything Now), prompts that ask the AI to roleplay as an unrestricted version of itself, prompts that hide instructions in encoded text.
Jailbreaks work because alignment training is shallow — the model learns to refuse certain patterns, but creative reformulations slip through. Each new model release tightens defences; researchers find new jailbreaks; cycle repeats.
On AskAI.free we don't condone jailbreaking, and most modern models (Claude Sonnet 4, ChatGPT 4o) are robust against well-known jailbreak prompts.
See it in action — ask any AI about jailbreak on AskAI.free.
Try it free →