Security
Studio security for agents: PII redaction, moderation categories, and blocking system-prompt override attempts. Applies to every outbound assistant path.
Security
The Security panel in Action Agent Studio stores an agent-level policy. ActionFlows applies it to outbound assistant text on Studio chat, the public API, SSE reconnect, and trigger replies. It is not a REST field you set per message.
PII redaction
When enabled, matching entities in assistant output are redacted before the text is returned.
Entities: email, phone, ssn, creditCard.
Moderation
When enabled, selected categories are checked. On a violation the agent either blocks the turn or warns (policy onViolation).
Categories: spam, profanity, nudity, hateSpeech, violence, selfHarm.
Prompt policy
blockSystemOverrideAttempts (on by default) rejects user messages that try to replace or ignore the system prompt.
What this is not
- It does not replace your product's auth. You still keep API keys on the server.
- It does not expose Trigger credentials.
- Public list/get agent responses still omit the security blob. Configure it in Studio.