ITG NowHome › AI glossary › Jailbreak
What is an AI jailbreak?
A jailbreak is a trick that gets an AI to ignore its safety rules and give answers it was built to refuse.
- Category Safety & law
- Also called jailbreaking
- Updated 20 Oct 2026
Jailbreak, explained simply
AI models have rules about what they won't do. A jailbreak is a clever prompt meant to get around those rules.
Companies test for jailbreaks and keep patching them. It is a back-and-forth between people finding gaps and makers closing them.
As a user, you don't need to try this. Jailbroken answers can be harmful or wrong, and trying may break a tool's terms of service.
On ITG Now, our daily AI news stories sometimes cover new jailbreak findings and how companies respond.
Related terms
Prompt injectionGuardrailsRed teamingAI safetyAI alignmentSystem prompt
Questions people ask about Jailbreak
Is jailbreaking an AI illegal?
It depends on where you are and what you do. It often breaks a service's terms, and harmful use can be illegal. This isn't legal advice.
Jailbreak vs prompt injection: what is the difference?
A jailbreak tries to make a model break its own rules. Prompt injection hides instructions in content so the model follows an attacker's wishes.
How can I stay safe from jailbreak risks?
Use trusted tools, keep them updated and double-check important answers. Don't paste in private data.