Jailbreak
Category: Agents“JAYL-brayk”
A trick that talks an AI out of its own safety rules, usually by roleplay, pressure, or a clever rewording of the request.
Lesson #9: Securing AI Agents: Cybersecurity Essentials →Talking a new security guard into letting you in because you 'work here' and 'left your badge at home'.
Someone asks a chatbot to 'pretend you're a character with no rules' to get an answer it would normally refuse.
A user convinces the site assistant to skip the safety checklist by framing it as a hypothetical training exercise.
A customer-facing bot is coaxed into revealing internal discount policy it was told never to share.
Safety rules written only in the prompt can be argued away, so real limits must be enforced in code, not in wording.
A jailbreak talks an AI past its safety rules using clever framing.
Try asking your company chatbot a borderline question two different ways. Note whether the answers match.
Hover or focus a choice to preview the new review date.
Exercises by category
- Agents0/21 · 0%
- Models0/10 · 0%
- Data0/11 · 0%
- Infrastructure0/22 · 0%
- Fundamentals0/15 · 0%
- NobleHub0/20 · 0%
Next badge: First Nail — 1 more exercise
Related terms
Picked automatically from the same category and the ideas this term shares.
Hand-picked pair
Deliberately attacking your own AI system to find weaknesses before an outsider does.
Hand-picked pair
An attack where hidden instructions are planted in a document, email, or web page so an AI reading it obeys the attacker instead of you.
Same category: Agents
Many AI agents working together while sharing information.
Same category: Agents
A group of very small AI programs that each have one job, but they work together like a team to solve bigger problems.
Same category: Agents
An AI agent that can complete tasks on its own within rules you give it.
Explained using this term
The instructions or question you give an AI so it knows what you want.