Glossary · Security

AI Guardrails

AI guardrails are the technical constraints placed around an AI system that limit what it can do regardless of what it is asked–in database tools, controls like read-only execution, query validation, schema-only context, and operation allowlists.

Guardrails are the answer to a simple question: what is the worst thing this AI can do? Without them, an AI connected to a database could in principle run any SQL–including destructive SQL–if it malfunctions or is manipulated. With guardrails, the ceiling is enforced by the system, not by the model's good behavior: writes are blocked at the permission level, generated queries are validated before execution, sensitive data never enters the AI's context, and prohibited operations are rejected outright. The principle is to treat the model as a capable but untrusted component, the way you treat user input. Guardrails are what make it reasonable to connect AI to production data at all.

Fig — every answer ships with the tables, rows and SQL behind it.