AI Safety

Why AI safety matters

Your AI assistants are talking to real customers on your behalf. Without guardrails, they can occasionally go off-script — discussing topics outside their scope, making statements you'd rather not have associated with your brand, or mishandling sensitive situations.

The AI Safety page (at AI Assistant > AI Safety) gives you the tools to prevent this. It's not about limiting your AI — it's about making sure it stays focused, appropriate, and trustworthy.

How Autoch.at's safety system works

Autoch.at uses a layered approach to safety:

1. Blocked keywords and topics — Hard stops that prevent the AI from engaging with specific content at all.
2. Serious Mode — A special handling mode for sensitive inquiries that require a more careful, measured response.
3. Handoff triggers — Automatic escalation to a human when the AI detects a situation it shouldn't handle alone.

These layers work together. A conversation might pass the keyword check but still trigger Serious Mode, which then escalates to a human.

Blocked keywords and topics

Blocked keywords are words or phrases that, if detected in a customer's message, will prevent the AI from responding normally. Instead, it will either:

• Redirect the conversation to a human agent.
• Send a pre-configured fallback message.
• End the conversation gracefully.

Adding a blocked keyword

1. Go to AI Assistant > AI Safety.
2. In the Blocked Keywords section, click Add Keyword.
3. Enter the word or phrase.
4. Choose the action to take when it's detected (redirect, fallback message, or end conversation).
5. Click Save.

Use this for topics that are completely off-limits — competitor names you don't want the AI to discuss, legally sensitive terms, or anything that should always go straight to a human.

Serious Mode

Serious Mode is triggered when the AI detects that a conversation involves a sensitive topic — things like mental health crises, legal threats, medical emergencies, or expressions of distress.

When Serious Mode activates, the AI shifts its behavior:

• It stops following its normal playbook and persona.
• It responds with calm, measured language focused on acknowledging the situation.
• It immediately escalates to a human agent or provides emergency resources, depending on your configuration.

Configuring Serious Mode

1. Go to AI Assistant > AI Safety.
2. In the Serious Mode section, toggle it on.
3. Choose the escalation action (route to a specific team, send an emergency message, or both).
4. Optionally, customize the message the AI sends when Serious Mode activates.

Reviewing safety events

Every time a safety guardrail fires — whether it's a blocked keyword or a Serious Mode activation — Autoch.at logs it. You can review these events in Insights > Analytics to understand how often your guardrails are being triggered and whether your configuration needs adjustment.

If you're seeing a lot of false positives (the AI escalating normal conversations unnecessarily), you may need to refine your blocked keyword list or adjust the sensitivity of Serious Mode.