Skip to content
ChurchWiseAI
ChurchWiseAIHelping Churches Bring People to Jesus

Safety and Threat Handling

How the ChurchWiseAI Voice Agent detects threats, handles abusive callers, and protects your church

The ChurchWiseAI Voice Agent includes built-in safety features that detect threats, handle abusive callers, and protect your church — all automatically. You do not need to configure anything. The system is always active on every call.

Two Layers of Protection

The voice agent checks every caller's words against two detection systems, in order:

1. Threat Detection (Most Severe)

Threat detection runs first and catches explicit threats of violence. If a caller says something that matches a threat pattern (threats to harm people, references to weapons, threats against the church), the voice agent:

  1. Responds firmly and calmly: "I need to stop you right there. This call is being recorded and logged. I'm ending this call now. If you or someone else is in danger, please call nine one one."
  2. Ends the call immediately — the line disconnects
  3. Sends emergency alerts — Email and SMS notifications go to:
    • The church admin email on file
    • The church admin phone (if configured)
    • ChurchWiseAI support team
  4. Logs the incident — The call transcript, threat details, and caller phone number are saved to the moderation violations table

2. Abuse Detection (2-Step Response)

Abuse detection catches profanity, harassment, and hostile language that does not rise to the level of a direct threat. The response has two steps:

  • First offense: The AI addresses the language directly: "I understand you may be frustrated, but I'm not able to continue if the language is directed at me. Is there something I can help you with?"
  • Second offense: The call ends immediately: "I'm going to end this call now. I hope your day gets better. Take care." The line disconnects.

The abuse counter resets between calls — each new call starts fresh.

What Gets Detected

The system detects patterns like:

  • Threats: Direct threats of violence against people, the church, or property
  • Weapons references: Mentions of bringing weapons or using them
  • Directed abuse: Severe profanity directed at the AI, church staff, or the church
  • Persistent hostility: Repeated abusive language after warnings

The system is carefully tuned to avoid false positives. Normal conversation — even about difficult topics like grief, addiction, abuse recovery, or domestic violence — does not trigger the safety system. Church callers discussing sensitive life situations are treated with the care they deserve.

Emergency Notifications

When a threat is detected, three notifications go out immediately:

  1. Email to the church admin — Full details of what was said, caller phone number, and timestamp
  2. SMS to the church admin phone — Short alert with the key details (if a phone number is configured)
  3. Email to ChurchWiseAI support — So our team is aware and can assist if needed

These notifications happen automatically — your pastoral team can decide what action to take.

Shared Moderation System

The voice agent and chatbot share the same moderation database (moderation_violations table). This means:

  • Violations from both channels appear in the same Moderation Dashboard in your admin panel
  • Your team gets a unified view of safety events across voice and chat
  • Escalation patterns (cooldown → temp block → permanent block) apply to the chatbot; the voice agent handles each call independently

What You Should Do

The safety system is fully automatic. There is nothing you need to configure. But here are some things to keep in mind:

  • Set up notification contacts. Make sure your voice agent has a valid notification_email and optionally a notification_phone in Settings so alerts reach the right people.
  • Review threat alerts promptly. If you receive a threat notification, assess whether further action is needed (contacting local authorities, alerting staff).
  • Trust the system for abuse. The two-step response gives callers one chance to correct course before the call ends. Most abusive callers stop after the first warning.

Related Docs