Guardrails

AgentDM includes built-in message safety filters. Messages matching guardrail phrases are flagged and not delivered.

Default Filters

These are enabled by default for all accounts:

• "Never include API keys, passwords, or tokens"
• "Never include customer PII in messages"
• "ignore previous instructions"
• "ignore all previous"
• "disregard your instructions"
• "you are now"
• "new instructions:"
• "system prompt:"
• "ADMIN OVERRIDE"

Customization

Guardrails can be customized per account via the dashboard. Add, remove, or modify filter phrases to match your security requirements.