Microsoft has been positioning Security Copilot as the AI layer across its security stack for the past year. In early 2026, that positioning got a lot more concrete for endpoint teams: Security Copilot agents are now embedded directly inside the Intune admin center, sitting in a dedicated Agents section and capable of running autonomous multi-step tasks without constant admin oversight.
This is different from the earlier “Copilot in Intune” experience, which was essentially a natural-language query layer on top of existing data. The new agents can take action: building remediation rings, flagging stale devices, generating policy configurations from raw compliance documents, and reviewing approval requests. Whether they’re ready for production use across the board is a different question, and we’ll get into that.
The Four Agents and What They Actually Do
Each agent handles a distinct part of the endpoint management lifecycle. They don’t replace your workflows but plug into them at decision points where context and scale normally create bottlenecks.
Vulnerability Remediation Agent: This one pulls data from Microsoft Defender Vulnerability Management, identifies CVEs affecting your managed device inventory, and autonomously builds a patching group with a proposed remediation plan. The key distinction is that the agent does not apply patches itself. It’s a planning and prioritization layer, not a push mechanism. Your admins still review and approve before anything deploys. Given how many orgs are sitting on sprawling unpatched inventories, this addresses a real pain point: turning a list of 3,000 CVEs into a prioritized remediation ring is exactly the kind of cognitive grunt work AI should absorb.
Policy Configuration Agent: Paste in a STIG, NIST guideline, or other compliance document and the agent parses it, maps requirements to Intune Settings Catalog entries, and generates a draft policy. This closes a gap that has frustrated endpoint admins for years. Compliance frameworks don’t ship with Intune templates, so the translation work has always fallen on someone who knows both the framework and the settings catalog well. The current version only accepts .txt files up to 100KB, which means you’ll need to extract and trim full CIS benchmark PDFs before uploading. More on that in the caveats section.
Device Offboarding Agent: Identifies devices that are stale, misaligned, or no longer active, and provides a pre-offboarding assessment before anything gets removed. The agent surfaces last check-in timestamps, compliance state, and assigned user so your decision to offboard is based on data, not guesswork from a manually filtered device list.
Change Review Agent: Evaluates requests sitting in Intune’s Multi Admin Approval queue and recommends approval or rejection based on the scope and risk of the requested change. This is particularly useful in enterprise environments where the approval queue accumulates requests faster than reviewers can assess them with full context.
Prerequisites and Licensing: What You Actually Need
Security Copilot agents in Intune run on Security Copilot’s backend, which means your environment needs to meet a specific bar before any of this is accessible.
Security Copilot must be provisioned first. You need to complete the initial setup in the Microsoft Security Copilot portal before any Intune integration activates. Agents in Intune show up after that first-run tour is completed. If your org hasn’t gone through this step, the Agents section in the Intune admin center will be present but inactive.
Licensing: Security Copilot is now included for Microsoft 365 E5 customers. You don’t need a separate Intune-specific add-on. Copilot licensing covers usage through Security Compute Units (SCUs). Agents consume SCUs the same way any other Security Copilot feature does. If you’re provisioned at pay-as-you-go SCU pricing rather than the full E5 inclusion, check your SCU consumption rates before enabling multiple agents in parallel. A large device inventory with frequent agent runs can add up.
Role requirements: Each agent has a minimum Intune role. The principle here is least privilege — you shouldn’t need Global Admin to activate a device offboarding assessment. Check the Microsoft Learn docs for agent-specific role requirements before assigning setup responsibilities.
Defender Vulnerability Management integration is specifically required for the Vulnerability Remediation Agent. Without an active MDVM license and integration configured, the agent has no data source to work from.
Setting Up the Vulnerability Remediation Agent: A Practical Walkthrough
The Vulnerability Remediation Agent is the most impactful starting point for most teams, so here’s how to get it running.
Step 1: Confirm prerequisites. Verify that Security Copilot is provisioned in your tenant and that Microsoft Defender Vulnerability Management is active and reporting data. Check that your Intune-managed devices are visible in the Defender portal.
Step 2: Navigate to Agents. In the Intune admin center, go to Agents in the left nav. You’ll see the available agents listed. Select Vulnerability Remediation Agent.
Step 3: Run the setup wizard. On the Overview tab, select Set up Agent. The wizard walks through current requirements: Defender integration status, SCU availability, and role confirmation. If anything is missing, the wizard surfaces it here.
Step 4: Start the first run. Select Start agent to kick off the initial scan. The agent reads current vulnerability data from MDVM, scopes it to your managed inventory, and begins building a remediation plan. First runs on large environments take time. This isn’t instant.
Step 5: Review the output. The agent presents a proposed patching group with CVE prioritization and recommended remediation actions. Review the groupings, verify the scope, and approve what you’re comfortable deploying. Nothing goes out until you explicitly authorize it.
Step 6: Set a cadence. After the first run, configure how often the agent re-scans. Weekly is a reasonable starting point for most orgs. Monthly will miss newly disclosed CVEs for too long; daily is overkill unless your threat model demands it.
Workflow Integration: Making Agents Part of Your Operations
The agents work best when they slot into existing approval and review processes rather than running independently. A few patterns worth considering:
For the Vulnerability Remediation Agent, align the agent’s output with your existing change management cadence. If you have a Tuesday patch window, configure the agent to complete its analysis by Monday so remediation groups are ready for same-week review and approval. Treat the agent’s output the same way you’d treat a report from a junior team member — it needs review, not blind trust.
For the Policy Configuration Agent, start with internal policy documents you already have translated into Intune before using it on external benchmarks. This lets you calibrate how well the agent handles your specific terminology and settings patterns before you’re relying on it for compliance work.
For the Change Review Agent, keep human review in the loop. Use the agent’s recommendations to triage the queue faster. High-confidence approvals where the agent flags low risk can move quicker; anything the agent flags as ambiguous gets full human attention.
Limitations and Caveats Worth Knowing
The Policy Configuration Agent has a hard file limit. Only .txt files up to 100KB are supported. A full CIS benchmark PDF, even converted to text, will exceed this. You’ll need to extract the relevant sections before uploading. Microsoft has indicated this limit may change, but currently it constrains how much of a benchmark document the agent can process in a single pass.
Agents don’t replace RBAC or change management. They introduce AI-generated recommendations into workflows, not AI-executed changes. Every meaningful action still requires human authorization. This is the right design choice, but it means the time savings are on the analysis and prioritization side, not on the execution side.
SCU consumption isn’t free. Even for E5 customers, Security Copilot has a provisioned SCU pool. Running multiple agents across large device inventories will consume SCUs at a higher rate. Build a baseline understanding of your SCU spend before enabling all four agents at once.
The Policy Configuration Agent deprecation note. As of early 2026, the current version of the Policy Configuration Agent has a transition period running through early June 2026. If you configure it before the cutoff, you can continue using it. After removal, the feature will go through an update cycle before returning. Worth tracking if policy configuration automation is part of your near-term plans.
Data residency and privacy. Security Copilot processes data through Microsoft’s AI infrastructure. Verify that your org’s data residency requirements are compatible with how Security Copilot handles device and vulnerability data before enabling agents in regulated environments.
Where This Is Headed
The Intune agent model is a preview of what endpoint management looks like when AI handles the pattern recognition and humans handle the decisions. The four agents released so far address the highest-friction points in the day-to-day workflow: CVE remediation at scale, compliance document translation, stale device cleanup, and approval queue triage.
Autonomous policy deployment, real-time threat response, and cross-tenant management at scale are still on the roadmap. For now, the practical value is in offloading analysis work — the agent reads 3,000 CVEs so you don’t have to, then asks you which ring to patch first.
For IT admins managing mid-to-large device inventories, the Vulnerability Remediation Agent alone is worth enabling today. The setup is straightforward once Security Copilot is provisioned, the output is actionable, and the control model keeps humans in the loop on actual changes. The rest of the agent suite is worth evaluating based on where your team’s specific bottlenecks sit. Not all four need to be active to get value from the integration.