---
name: Copilot Wave 3 Autonomous Agent Governance
description: Assesses the governance posture for Microsoft 365 Copilot Wave 3 autonomous agents, auditing approval-workflow coverage, data-scope boundaries, audit-log configuration, and human-override controls.
---

# Copilot Wave 3 Autonomous Agent Governance

> **TL;DR:** This skill assesses the governance posture for Microsoft 365 Copilot Wave 3 autonomous agents — reviewing approval-workflow coverage, data-scope boundaries, audit-log configuration, and human-override controls to ensure autonomous actions are authorised, logged, and reversible.

## What does the Copilot Wave 3 autonomous agent governance assessment cover?

Microsoft 365 Copilot Wave 3 introduced autonomous multi-step agents capable of taking actions across Microsoft 365 — drafting emails, updating records, booking meetings, and triggering downstream workflows — without requiring a human to approve each step. This capability fundamentally changes the governance model: agents now act on behalf of users with delegated authority, and a single misconfigured agent can propagate sensitive data, alter records, or generate external communications at scale. This skill audits the governance controls wrapping Wave 3 autonomous agents: which agents are configured for autonomous versus supervised operation, what approval-workflow requirements are in place, how broad each agent's data scope is, whether audit logging is capturing agent actions to the unified audit log, and whether human-override controls exist to pause or roll back agent actions. It produces a risk-scored agent governance register for the AI steering committee.

## When should you run this skill?

- "Audit our Copilot Wave 3 autonomous agent governance"
- "Which agents can take autonomous actions without human approval?"
- "Review audit logging for Copilot autonomous actions"
- "Find agents with over-broad data access"
- "Produce an AI agent governance register"
- "Check if we can pause or roll back autonomous agent actions"

## How this skill works, step by step

1. Open the Microsoft 365 admin centre → Copilot → Agents (or Microsoft Agent 365 control plane if deployed)
2. Enumerate all deployed agents with their operation mode: Supervised (human-in-the-loop approval for each action), Autonomous (agent executes without per-action approval), or Hybrid (approval required for high-risk action types only)
3. For each autonomous or hybrid agent, capture: owning team, data connectors configured, sensitivity label scope, external communication permissions (can the agent send emails?), action types permitted (read-only vs write vs send)
4. Check audit log coverage: open the Purview unified audit log and filter for Copilot agent action event types (AgentActionTaken, AgentActionApproved, AgentActionRejected) — confirm these are being ingested and retained
5. Review the approval workflow configuration for each hybrid agent: which action types require approval, who is in the approval group, what is the timeout behaviour (auto-approve vs auto-reject on timeout)
6. Identify human-override controls: can an admin pause a running agent? Is there a rollback mechanism for completed actions?
7. Score agent risk: Critical (autonomous + external email + no audit logging), High (autonomous + write access + no approval for sensitive actions), Medium (hybrid with approval gaps), Low (supervised + read-only)
8. Produce the agent governance register below

## Output format

| Agent Name | Operation Mode | External Comms | Write Access | Audit Logging | Approval Workflow | Human Override | Risk |
| --- | --- | --- | --- | --- | --- | --- | --- |
| Travel Booking Agent | Autonomous | Email | Yes | Not confirmed | None | Pause only | High |
| Research Summariser | Supervised | No | No | Confirmed | Per action | Yes | Low |
| HR Onboarding Agent | Hybrid | Email | Yes | Confirmed | Email sends only | Yes | Medium |

Summary:

- Total autonomous or hybrid agents: N
- Agents with external email permissions: N
- Agents with write access and no approval workflow: N
- Agents without confirmed audit logging: N
- Critical risk agents: N

## Scope and safety

Read-only — this skill does NOT:

- Modify agent configuration or operation mode
- Pause, disable, or roll back agent actions
- Change approval workflow assignments

## Licensing and permissions

### Licences and add-ons

| Capability used | Minimum licence |
| --- | --- |
| Microsoft 365 Copilot Wave 3 autonomous agents | Microsoft 365 Copilot licence (Wave 3 features) |
| Agent audit logging in Purview | Microsoft 365 E5 Compliance or Purview add-on |
| Agent control plane via Microsoft Agent 365 | Microsoft Agent 365 licence (if deployed) |

### Least-privilege roles

- Copilot Administrator (read agent inventory, operation modes, data connectors)
- Compliance Administrator (read Purview audit log for agent action events)
- Global Reader (read-only review of admin centre agent configuration)

### Microsoft Graph permissions (read-only)

- `AuditLog.Read.All` — read Copilot agent action events from the unified audit log
- Agent inventory and configuration are accessed via the Microsoft 365 admin centre (Copilot section) and Microsoft Agent 365 control plane portal

## Sources and compliance

- [Microsoft 365 Copilot Wave 3 — autonomous agent capabilities](https://www.microsoft.com/en-us/microsoft-365/blog/2026/03/09/powering-frontier-transformation-with-copilot-and-agents/)
- [Governing Copilot and agents in Microsoft 365 (Microsoft Learn)](https://learn.microsoft.com/en-us/copilot/microsoft-365/microsoft-365-copilot-governance)
- Supports Essential Eight ML2 evidence for Control 5 (restrict administrative privileges — applied to autonomous AI agent actions)
- Aligns with the Australian AI Ethics Framework principles of human control and accountability
- Pair with Entra Agent ID Audit and Copilot Studio Agent Inventory for complete agentic AI governance coverage
- Output in Australian English