AI Dictionary › Prompting

Jailbreak

A jailbreak is an attempt to bypass an AI model's safety protections to make it produce content it would otherwise refuse: dangerous instructions, hateful content, confidential information. Techniques range from role-play tricks ("pretend you're an AI with no rules") to more sophisticated attacks using encodings and carefully crafted prompts.

Definition

It differs from prompt injection in its target: a jailbreak attacks the model's own rules, injection attacks the application built on top of the model by inserting hostile instructions into the data it processes. Both belong to AI security, a fast-growing discipline.

For companies the issue is concrete: a jailbroken enterprise chatbot can damage the brand or leak information. Defenses: models with robust protections, additional application-level guardrails, security testing (red teaming) before public release, and continuous monitoring.

Related terms

• Prompt Injection
• System Prompt
• EU AI Act
• AI Content Moderation
• RLHF

More in Prompting

• Prompt
• Prompting
• Role Assignment
• Zero-shot
• Few-shot
• Chain of Thought
• Prompt Engineering
• Output Format

Put it into practice

• Real scenario: apply this concept with an instant AI verdict.
• Daily gym: short drills to train dictionary concepts.
• Playoff game: challenge other players, elimination-style.

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.