Safety guardrails | Civio Civio Assist Knowledge Base     [Skip to content](#main)  [ ![Civio Civio Assist Knowledge Base](/civio-light.svg) ](https://nginx.deploy-lagoon-production.civio-assist-docs.dh1.amazee.io)  [Product documentation](/articles) [Release notes](/releases) [ Support portal ](https://nginx.deploy-lagoon-production.civio-assist-docs.dh1.amazee.io/portal/login)   Toggle navigation      

  [Product documentation](/articles) [Release notes](/releases) [Support portal](https://nginx.deploy-lagoon-production.civio-assist-docs.dh1.amazee.io/portal/login)  

  1. [Home](/) ›
2. [Product documentation](/articles) ›
3. [Privacy and safety](/articles?category=4) ›
4. Safety guardrails

 Safety guardrails
=================

Updated 1 month ago · 2 min read

 Search     Search  

Sometimes the chatbot deliberately does not answer. When that happens you will see a **guardrail** marked against the message in conversation history.

A guardrail block is usually the chatbot working as intended: it is declining a question it was never meant to answer. There are two kinds, and they differ in what they protect and how you review them.

Bot scope guardrail
-------------------

The question fell outside the topics this chatbot is set up to answer. The user receives the fallback message configured for your bot rather than a normal reply.

Most of these blocks are correct: the question really was outside what the chatbot covers. Occasionally a blocked question does belong in scope, and the knowledge base was missing the information to answer it.

These messages carry the **Update knowledge base** action. If, on review, you decide the question belongs in scope, it proposes the missing content without leaving the conversation. See [Suggesting knowledge straight from a conversation](/articles/manual-context-from-a-conversation).

![A conversation thread showing a message stopped by a bot scope guardrail, with the Update knowledge base action available on it](https://assist.knowledge.civ.io/storage/33XXGFKgrRLkkfhswx1ZGSWBUvxnSUSUfHmovqXA.png)

Platform safety guardrail
-------------------------

The message was blocked by the platform's safety filters. These catch things like hate speech, self-harm, sexual content, violence, and attempts to manipulate the chatbot into ignoring its instructions.

There is no inline action on these, the filters apply to every chatbot on the platform.

Telling them apart
------------------

Bot scope

Platform safety

Why it stopped

Outside the topics this bot covers

Blocked by the platform's safety filters

What the user saw

Your configured fallback message

Your configured fallback message

What to consider

Whether the question belongs in the chatbot's scope

Nothing to change; the filters apply to every chatbot

Reviewing guardrails
--------------------

Filtering conversation history to **With guardrail blocks** is the fastest way to review the questions your chatbot declined. Expect most to be genuinely off-topic; the occasional one that belongs in scope is worth proposing for the knowledge base.

See [Finding a conversation](/articles/finding-a-conversation) and [Reading a conversation and understanding alerts](/articles/reading-a-conversation-and-alerts).

A steady number of guardrails is healthy. It means the chatbot is staying inside its remit instead of guessing at questions it was never set up to handle.

Was this article helpful?
-------------------------

Your feedback helps us prioritise what to rewrite.

      Yes     No  

 [    Back to Privacy and safety ](/articles?category=4) 

 On this page 

### Still need help?

Raise a request in the support portal and track it to resolution.

 [ Submit a support request ](https://nginx.deploy-lagoon-production.civio-assist-docs.dh1.amazee.io/portal/login)

**Need additional help?** Contact our support team at [help@assist-support.civ.io](mailto:support@civ.io)

 © 2026 [Civio](https://civ.io)