Skip to content
← Security & Compliance

Security & Compliance

Auditor Skill & Human-in-the-Loop

Real-time PII, hack, and custom auditors plus the Escalation Skill and Copilot takeover workflow.

RBAC secures who can build and manage agents. The Auditor and Escalation skills secure what agents and users actually say — providing automated compliance monitoring and human-in-the-loop intervention at runtime.

01

The Auditor Skill

A real-time security firewall for AI conversations. The Auditor scans every incoming and outgoing message for security risks, privacy violations, and compliance breaches.

Auditor skill flow
ModuleDetects
PII AuditorEmail, phone, SSN, financial data, and other Personally Identifiable Information.
Hack AuditorPrompt-injection attempts, system abuse, and instructions designed to bypass the agent.
Custom AuditorOrganization-defined rules using custom prompts and keywords (e.g., flag medical-advice requests).
02

Alert Dispatch & Monitoring

When a module detects a violation, it normalizes the event, logs it securely, and dispatches an alert via email or webhook(ideal for forwarding to SIEM platforms like Splunk or to Slack). All triggered events appear in the centralized Alerts Dashboard, filterable by auditor type and date range.

03

Escalation Skill

Automation is powerful, but some scenarios require human empathy and judgment. The Escalation Skill bridges AI efficiency and human oversight.

Escalation and HITL flow
  1. The agent watches the conversation against escalation triggers (e.g., "urgent", "cancel account", "speak to a human").
  2. On trigger, the skill sends an email, SMS, or webhook alert to the designated team, including the full conversation payload and a custom internal prompt.
  3. The AI replies with a predefined message acknowledging the handoff (e.g., "I've alerted our billing team — a specialist will contact you shortly.").
04

Human Takeover via Copilot (HITL)

Take Over Conversation

An Admin clicks Take Over Conversation. The AI is immediately disabled for that thread. The Admin types responses manually through Copilot — messages appear with their name and an admin tag for transparency. When the issue is resolved, Return to AI hands the conversation back.

Regular users and editors cannot view all conversations or take over threads. Only users with the Agent Admin role (or higher) can enable Admin Mode in Copilot, which grants full visibility into historical and active chat logs across SMS, Email, Voice, and Live Chat.