# What is guardrails?

> LLM guardrails are runtime controls that detect or block unsafe prompts and responses—such as prompt injection, sensitive topics, and topic restriction—before they reach users or tools.

- HTML: https://openlit.io/glossary/guardrails
- Markdown: https://openlit.io/glossary/guardrails.md

Guardrails sit in the governance layer of the harness. They fail closed for safety-critical paths and should be traced so you can see when and why a request was blocked.

OpenLIT includes open-source guardrails for prompt injection and related checks, correlated with OpenTelemetry traces so blocked calls show up next to the agent step that triggered them.


## Related

- [Agent harness](https://openlit.io/glossary/agent-harness.md)
- [Agent evals](https://openlit.io/glossary/agent-evals.md)
- [Harness engineering](https://openlit.io/glossary/harness-engineering.md)
- Pillar: https://openlit.io/agent-harness-engineering.md