AI Dictionary › Prompting

Guardrails

Guardrails are the technical controls that bound an AI system's behavior: input filters (blocking inappropriate requests or injection attempts before they reach the model), output constraints (verifying responses respect policy, format and allowed topics) and behavior rules (forbidden subjects, mandatory escalation to humans in sensitive cases).

Definition

They are the application layer above the model's native protections: a banking chatbot must not give investment advice, an HR assistant must not comment on individual employees, and it's not enough to hope the model figures this out on its own, it has to be enforced with explicit controls.

Dedicated tools (from open-source libraries like Guardrails AI and NeMo Guardrails to cloud provider services) let you define these rules declaratively and verifiably. In serious enterprise AI systems, guardrails are part of the design from day one, they are privacy-by-design applied to model behavior.

Related terms

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

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.