Tricking AI into breaking rules
Some people trick AI into breaking its rules to do things it normally refuses. That trick is called a jailbreak.
TRICKS
Role-play and roundabout stories
They often make AI play a role, tell a roundabout story, or split a harmful request into many small steps.
DANGER
Why it's dangerous
The danger is that AI could give harmful instructions, and people who try it can get their accounts banned.
DEFENSES
Many layers of guardrails
AI companies block it in layers: training the model to spot tricks, adding filters, and patching new holes quickly.
This article is based on the video Jailbreak from the Mark học AI channel. Watch the video (in Vietnamese) to see the animations.