AI Safety
Keep your AI on-brand and careful. Add words it must never say, and words that make it switch to a more serious, empathetic tone. Admin-only setting.
AI assistants talk to real customers on your behalf. AI Safety gives you two simple lists to keep them on-brand and careful:
- Blocked Keywords: words or phrases the AI must never use in a reply.
- Serious Mode Keywords: words in a customer's message that tell the AI the topic is serious, so it answers in a more professional, empathetic and careful way.
Open AI Assistant → AI Safety in the sidebar.
Only organization admins can open this page. Other team members see "Access Denied."
How the safety net works
Autoch.at combines two sets of keywords:
- System boundaries: built-in keywords that are always on. You can see them under each box, but you can't remove them.
- Your keywords: the ones you add.
When they combine, the stricter rule wins. If the system blocks something, it stays blocked even if it isn't in your list. Under each box, Effective (combined) shows the full list actually being used.
Blocked Keywords
Use this for words or phrases your AI should never say, such as a competitor's name, a confidential project name, or a term you don't want linked to your brand. If the AI tries to use one in a reply, that reply is blocked automatically.
Open AI Safety
Go to AI Assistant → AI Safety.
Add your words
In Blocked Keywords, type one keyword or phrase per line.
Save
Click Save Settings at the bottom. You'll see "AI safety keywords saved successfully."
Serious Mode Keywords
Some conversations need extra care: legal threats, fraud, safety concerns, formal complaints. In Serious Mode the AI becomes more professional, empathetic and careful.
Autoch.at already watches for a built-in set of serious topics. You can add your own in two lists:
| List | What happens | Example words |
|---|---|---|
| High Confidence Keywords | Triggers Serious Mode immediately. | lawsuit, legal action, fraud, theft |
| Medium Confidence Keywords | Autoch.at double-checks with AI before switching. | very upset, extremely frustrated, privacy violation |
Type one keyword or phrase per line, then click Save Settings.
Start small. A long list of common words can switch the AI into Serious Mode too often. Check real conversations in Conversations to see if the tone is right, then adjust your lists.
If the page won't load
You'll see "Couldn't load your safety settings." Your saved keywords haven't changed. Click Retry. Autoch.at will not let you save while the page failed to load, so you can't accidentally wipe your lists.
Safety is more than keywords
| You want to... | Go to |
|---|---|
| Set how the AI should sound everywhere | Messaging Directives |
| Give an assistant "never do" rules | AI Assistants (Do Not Instructions, Topic Boundaries) |
| Control when a person takes over | Routing & Handoffs |
| Control what personal info is collected or kept | PII & Privacy |