Policies are the guardrails around what an AI Employee may do. The system is default-deny: an action is allowed only if the employee has the tool ticked and no policy blocks it.
What you control:
- Tool permissions per employee (Settings → AI Employees → Edit).
- Autonomy level, which decides whether allowed actions run or wait for approval.
- Policy verdicts you see in Reports → AI Activity: allowed, requires human (goes to AI Approvals) or blocked, each with a reason.
- Contact-level rules such as STOP opt-outs and quiet hours (no automated messages between 9 pm and 8 am) are always enforced.
Every decision is stored with WHAT (the proposed action), WHY, GUARDRAIL (the verdict) and RESULT (executed, blocked, awaiting a human, rejected). Use it to audit or to debug why something did not happen.