Skip to content

Guard model

A guard model is a small AI that checks questions and answers for dangerous or offensive content. Domino IQ can use a guard model from Notes 2026.

A guard model is a small language model that stands guard. It reads the question before it reaches the large model. It can also read the answer before the user sees it.

What does it stop?

It stops content that is dangerous, unsafe or offensive. A question about building a bomb is a typical example. It can also catch attempts to trick the AI into breaking its rules.

A guard model does not give full security. It also costs a little extra computing power. But it is a good layer on top of the others.

Guard model in Domino IQ

From Notes 2026 Domino IQ can use a guard model. The admin chooses it in dominoiq.nsf. Domino IQ can also rewrite the user's text before it is sent on. That method is older, and HCLSoftware now recommends turning it off and using a guard model instead.

All terms in the glossary