Home › Key concepts › Checks before and after

Key concept · Walk (intermediate) · the same on every platform

Checks before and after

Code that inspects what goes into the model and what comes out, and can stop it.

What it is

Input checks look for things like personal data or attempts to override instructions. Pre-tool checks confirm a requested action is allowed. Output checks confirm the answer is grounded, on-topic and makes no forbidden promise. Many are plain code; some use a smaller model.

Why it matters

A model can be fooled. A check that runs every time does not get fooled the same way.

An example

The model is tricked into asking for a payout; a pre-tool check refuses because the customer never asked for one.

The common mistake

Trusting a line in the prompt ('never pay without approval') as if it were a control.

Where you meet it on each platform

The idea stays the same; only the product name changes.

JobAWSAzureGoogle CloudDatabricksOpen source
Guardrailsblock what must never go in or come outBedrock GuardrailsAzure AI Content SafetyModel ArmorAI Gateway guardrailsrule checks in your own code

Related ideas

Where it shows up

Part of the Walk level on these paths: AI Engineering (safety and proof).