Intune Change Review Agent: Would You Approve It?
Would you approve a PowerShell script because an AI labeled it low risk? The Intune Change Review Agent can surface Defender, Entra, and Intune signals—but it cannot see your rollout plan, rollback proof, or business context.
In public preview, the agent evaluates Windows PowerShell script requests protected by Multi Admin Approval. It can make a review faster, but it does not own the decision: a separate administrator must still approve, reject, or ask for more context.
The Problem
A script review is more than malware scanning. The approver needs to understand who requested the change, what the code does, which devices receive it, what similar changes did before, and how the team will recover if the outcome is wrong.
That context is easy to miss when an urgent request arrives with a weak business justification. Multi Admin Approval adds separation of duties; the agent adds structured risk evidence.
The Nugget
Use five checks before accepting an Intune Change Review Agent recommendation:
Check Approval evidence
Intent Clear business outcome and complete script source
Identity Requestor role, Entra risk, and separation from the approver
Technical risk Agent factors, Defender signals, external calls, and privileged actions
Blast radius Assignment groups, exclusions, device count, and test-ring result
Recovery Detection, rollback, owner, and post-deployment verification
The agent can recommend Approve, Reject, or Needs more info. Those labels are decision support. Microsoft explicitly keeps the final approval or rejection with the administrator.
Why This Matters
The current preview scope is narrow: Multi Admin Approval requests for Windows PowerShell scripts. The agent uses Defender Vulnerability Management for threat insights, Entra ID for identity risk, and Intune for request history and context.
Licensing and plugins are part of the control path. Microsoft documents Intune Plan 1, Entra ID P2, Defender Vulnerability Management, Security Copilot capacity, and the required Intune, Entra, Defender XDR, and Threat Intelligence plugins.
Operational details matter too. Runs are manual, not scheduled, and current documentation limits a run to ten requests. If the assigned identity is unused for 90 consecutive days, authentication can expire. Build ownership for running and renewing the agent instead of assuming it is a continuous control.
Multi Admin Approval also has a lifecycle. The requestor cannot approve their own change. After approval, the original requestor must complete the change, and an unprocessed request expires after three days.
What Admins Should Do
-
Protect Windows PowerShell scripts with a dedicated Multi Admin Approval access policy.
-
Use separate requestor and approver groups with least privilege.
-
Require script source, target scope, test evidence, and rollback in the justification.
-
Run the agent and compare its factors with a manual peer review.
-
Record why the admin accepted or overruled the recommendation.
-
Complete the change, then verify deployment status and audit records.
Pilot with known low-risk and deliberately suspicious test scripts. The goal is not to measure whether the agent agrees with the team; it is to learn which signals it surfaces and which context remains human-only.
Verification Path
For each pilot request, preserve the script hash, requestor, business justification, agent recommendation, approver note, final decision, completion status, and device outcome. This produces an auditable evidence chain instead of a screenshot of an AI label.
Pro Tip
Make a successful test-ring result mandatory for broad assignments. The agent can assess risk signals, but only your environment can prove that a script behaves correctly on your managed devices.