Safety & Ethics
Glossary
Guardrails
Rules and checks that keep an AI system within safe, intended behaviour.
Last updated
Guardrails are the constraints around a model: system prompts, input and output filters, permission checks and human approval for risky actions. They matter most in agents that can act on the world. Good guardrails are layered, because any single check can be bypassed, and they should fail safely rather than silently.
Related terms
System Prompt
High-level instructions that set a model's role, tone and rules before any user message.
Hallucination
When a model confidently states something that is false or unsupported by its sources.
Bias
Systematic unfairness in a model's outputs, usually inherited from its training data or design choices.
AI Agent
A system that uses a model to plan and carry out multi-step tasks, often calling tools along the way.
Keep exploring
Browse the full AI glossary or compare AI tools that use this technology.