Microsoft Defender for Cloud is switched on in most Azure tenants, yet many teams run only the free tier, chase secure score points without regard to risk, and never connect its alerts to anyone who will act on them. It is Microsoft's cloud-native application protection platform (CNAPP), combining posture management, workload protection and DevOps security across Azure, AWS, GCP and on-premises servers. This guide explains how to choose between Foundational CSPM and Defender CSPM, which workload protection plans matter for which estates, how the identities and roles behind the service work, how to route its findings, and how to read secure score without being misled by it. It is for cloud security engineers, SOC leads and platform owners. For the CNAPP category in general see the CNAPP Guide; this page covers Defender for Cloud specifically.
This is an independent guide from NHI Mgmt Group, based on Microsoft's public documentation as of November 2026. It is not sponsored or reviewed by Microsoft. Product features change, so confirm settings against Microsoft's current documentation.
Key takeaways
- Foundational CSPM is free but shallow. It gives recommendations against the Microsoft cloud security benchmark, secure score and AWS and GCP connectors; attack path analysis, regulatory compliance, agentless scanning, CIEM and data posture need the paid Defender CSPM plan.
- Workload plans are chosen per resource type. Servers, Storage, Databases, Containers, App Service, Key Vault, Resource Manager, APIs and AI services are separate plans; enable the ones that match where your sensitive data and attack surface sit.
- The connectors are privileged non-human identities. Onboarding AWS and GCP creates roles that can read resources, snapshot disks and, for data scanning, read buckets; treat them like any other high-privilege integration.
- Secure score measures coverage, not safety. The classic score awards a control's points only when every recommendation is fixed on every resource; the newer risk-based Cloud Secure Score in the Defender portal weights by risk and asset criticality. Do not compare them.
- Value comes from routing. Continuous export, workflow automation and governance rules turn findings into owned work; without them alerts go to subscription owners' inboxes and age.
How identity and access work in Microsoft Defender for Cloud
Defender for Cloud uses Azure RBAC at subscription, resource group and management group scope. Two security-specific built-in roles sit alongside the usual Reader, Contributor and Owner, and the table below, from Microsoft's permissions reference, shows why the choice matters.
| Action | Security Reader or Reader | Security Admin | Contributor (subscription) | Owner (subscription) |
|---|---|---|---|---|
| View alerts and recommendations | Yes | Yes | Yes | Yes |
| Apply a recommendation fix | No | No | Yes | Yes |
| Dismiss alerts | No | Yes | Yes | Yes |
| Enable or disable Defender plans | No | Yes | Yes | Yes |
| Assign initiatives, including regulatory standards | No | Yes | No | Yes |
| Edit security policy | No | Yes | No | Yes |
| Exempt recommendations | No | Yes | No | Yes |
- Security Admin can hide risk: it can dismiss alerts, exempt recommendations and turn plans off, without being able to fix anything. That combination is exactly what an attacker or a tired analyst would use to make problems disappear, so it belongs in Privileged Identity Management with approval, not as a standing assignment.
- Remediators need resource rights: the Fix button needs Contributor-level rights on the resource. Give remediation to the resource owners, not to the security team's Security Admin role.
- Owner for full plan capability: Microsoft notes that enabling all capabilities of a plan requires Owner, because some plans create role assignments for their own managed identities.
- SOC access: Security Reader at management group scope is usually enough for analysts to triage across the estate.
Defender for Cloud is also moving into the Microsoft Defender portal, where some features, including the risk-based Cloud Secure Score, now live. Access there is governed by Defender portal permissions as well as Azure RBAC, so review both when you set up the SOC. See the Azure RBAC and Service Principals Guide.
Choosing CSPM and workload protection plans
Foundational CSPM or Defender CSPM
| Capability | Foundational CSPM (free) | Defender CSPM (paid) |
|---|---|---|
| Asset inventory, recommendations, MCSB, secure score, workflow automation | Yes | Yes |
| AWS and GCP agentless connectors; DevOps connectors for Azure DevOps, GitHub and GitLab | Yes | Yes |
| Attack path analysis and cloud security explorer | No | Yes |
| Regulatory compliance assessments | No | Yes |
| Agentless VM vulnerability and secrets scanning | No | Yes |
| Permissions management (CIEM) across Azure, AWS and GCP | No | Yes |
| Data security posture management and AI security posture management | No | Yes |
| Governance rules, custom recommendations, critical asset tagging, internet exposure analysis, EASM | No | Yes |
| Code-to-cloud mapping and pull request annotations | No | Yes |
Defender CSPM is billed per resource of specific types (for example virtual machines, storage accounts, SQL servers, function and web apps in Azure; EC2, S3 and RDS in AWS), not a flat fee, and Microsoft added billing for serverless resources during 2026. Choose it when you need prioritisation rather than a flat list: attack paths that combine internet exposure, a vulnerable workload and an identity with access to sensitive data are the findings that justify the cost. A small single-subscription estate with a short recommendation list can often manage with Foundational CSPM plus a few workload plans; a multi-cloud or regulated estate will usually need Defender CSPM for compliance reporting and CIEM.
Workload protection plans
| Plan | What it adds | Enable when |
|---|---|---|
| Defender for Servers Plan 1 | Defender for Endpoint EDR integration; can be enabled per resource | You need EDR on cloud and Arc-connected servers and little else |
| Defender for Servers Plan 2 | Plan 1 plus agentless scanning (vulnerabilities, malware, secrets), premium Defender Vulnerability Management, OS baseline assessment against MCSB, file integrity monitoring, just-in-time machine access, data ingestion benefit; DNS alerts for new subscriptions | Servers are a meaningful part of the estate; subscription-level only |
| Defender for Storage | Activity anomaly detection, SAS misuse detection, malware scanning per GB, sensitive data threat detection | Any account holding sensitive or uploaded content |
| Defender for Databases | Threat detection for Azure SQL, SQL on machines, open-source relational databases, Cosmos DB | Production data stores |
| Defender for Containers | Kubernetes hardening, image vulnerability assessment, runtime protection | AKS, EKS or GKE in production |
| Defender for Key Vault | Unusual access to secrets, keys and certificates | Every subscription with vaults |
| Defender for Resource Manager | Suspicious management operations, known Azure attack toolkits, lateral movement from management to data plane | Every subscription: this is the control plane |
| Defender for App Service, APIs, AI services | Threats to web apps, published APIs and generative AI applications | Where those workloads exist |
For the server plans in depth see the Microsoft Defender for Endpoint Guide.
Admin roles and privileged access
Defender for Cloud itself creates identities and touches privileged settings, and attackers aim at it because disabling it blinds the defender.
- Plan and policy changes: disabling a Defender plan, removing the MCSB initiative assignment or deleting continuous export settings are defence impairment actions (ATT&CK v20 tactic TA0112 Defense Impairment, technique T1685 Disable or Modify Tools). Alert on them in the Azure Activity log and enforce plan enablement with Azure Policy.
- Exemptions: an exemption marked Risk accepted (waiver) removes resources from the score; Resolved through third-party service (mitigated) counts them as healthy and raises the score. Both need Owner, Security Admin or Resource Policy Contributor, can be set at management group scope where they cover future resources too, and take up to 24 hours to apply. Require an expiry date and a ticket for every one.
- Auto-provisioning identities: when Security Admin configures agents and extensions through policy remediation, Defender for Cloud creates managed identities with subscription-level roles; Defender for Containers, for example, uses Kubernetes Extension Contributor, Contributor, Azure Kubernetes Service Contributor and Log Analytics Contributor. Inventory these and do not reuse them.
- Just-in-time machine access: Plan 2 can open management ports on request and close them after a set time. Restrict who can request access and log every request.
See the Privileged Identity Management (PIM) Guide for eligible assignments.
Integrations, API tokens and service accounts
Multicloud and DevOps coverage depends on connector identities that Defender for Cloud creates in other environments. These are long-lived non-human identities with wide read access, and some with write access.
| Environment | Identity created (examples) | Access | What to check |
|---|---|---|---|
| AWS (Defender CSPM) | CspmMonitorAws | Read resources for discovery | Trust policy limited to the documented Microsoft principal; no edits to the policy |
| AWS (agentless scanning) | DefenderForCloud-AgentlessScanner | Create and clean up tagged disk snapshots, manage KMS keys | CloudTrail for snapshot and KMS activity outside the expected pattern |
| AWS (data posture) | SensitiveDataDiscovery | S3 read and KMS decrypt | Effectively data access: scope and monitor it |
| AWS (CIEM) | DefenderForCloud-Ciem, DefenderForCloud-OidcCiem | Discovery via sts:AssumeRole | Role session activity in CloudTrail |
| AWS (Servers) | DefenderForCloud-DefenderForServers, DefenderForCloud-ArcAutoProvisioning | Security group changes for JIT; Arc install through SSM | Write access: treat as privileged |
| GCP | MDCCspmCustomRole, MDCAgentlessScanningRole, data-security-posture-storage, microsoft-defender-ciem | Discovery, disk snapshots, bucket object read, IAM policy read | Cloud Audit Logs for the service accounts |
- DevOps connectors: connecting Azure DevOps, GitHub or GitLab grants Defender for Cloud access to repositories to find IaC misconfigurations and exposed secrets. Use the narrowest organisation scope and review the app's permissions in the code host.
- Workflow automation: Logic Apps triggered by alerts, recommendations or compliance changes run under their own connections and identities, and Teams, Outlook or ticketing connectors may hold credentials. Creating them needs Logic App Contributor plus Security Admin or Owner on the resource group.
- CIEM findings are NHI findings: the CIEM capability in Defender CSPM covers Entra service principals, AWS IAM roles and GCP service accounts, recommending removal of inactive identities and right-sizing over-provisioned ones. Microsoft states the retirement of Entra Permissions Management does not affect it.
A common attack pattern shows why this matters: an attacker holding a compromised subscription owner account creates an application, grants it Contributor on a subscription, adds credentials to existing apps and runs cryptomining virtual machines under those application identities. Resource Manager telemetry, service principal role assignment review and CIEM findings are where that pattern surfaces. See the Entra ID Applications and Consent Guide.
Logging and detection
Alerts and recommendations are only useful once they reach a queue with an owner. Defender for Cloud provides three routes.
- Continuous export: streams alerts, recommendations and, optionally, vulnerability findings to a Log Analytics workspace or Event Hubs, in another subscription or tenant if needed. Streaming sends changes in health state; weekly snapshots (flagged
IsSnapshot) give a full picture for reporting. Deploy at scale with the DeployIfNotExist policy. - SIEM and XDR: send alerts to Microsoft Sentinel or another SIEM and correlate with Entra sign-in and Activity log data. See the Microsoft Sentinel Guide.
- Workflow automation and governance rules: Logic Apps for enrichment and ticketing; governance rules (Defender CSPM) assign recommendations to owners with due dates at scale.
| Detection source | Example behaviour | ATT&CK v20 |
|---|---|---|
| Defender for Resource Manager | Operations from malicious IPs, known Azure attack toolkits, suspicious VM extension scripts | T1651 Cloud Administration Command; T1580 Cloud Infrastructure Discovery |
| Defender for Resource Manager | Antimalware disabled through the management plane | T1685 Disable or Modify Tools |
| Activity log plus CIEM | New Contributor or Owner assignment to a service principal | T1098.003 Additional Cloud Roles |
| Defender for Key Vault, Storage | Unusual identity reading secrets or bulk-reading blobs | T1078.004 Cloud Accounts |
Reading secure score properly
The classic secure score in the Azure portal counts only built-in, generally available MCSB recommendations, is recalculated every eight hours per subscription or connector, and gives a control's maximum points only when all its recommendations are fixed on all resources. Since 30 June 2026 AWS and GCP recommendations also count, so a drop may reflect wider coverage rather than worse posture. The Cloud Secure Score in the Defender portal (Exposure Management) weights open recommendations by risk level, adjusted for internet exposure, data sensitivity and asset criticality. Use the risk-based score for prioritisation and the classic score for control coverage, and report trends rather than single values.
Metrics
- Plan coverage: percentage of subscriptions and connectors with the agreed plans enabled.
- Critical attack paths: open attack paths to critical assets, and median age.
- Recommendation ageing: high and critical recommendations past their governance due date.
- Exemption hygiene: exemptions without an expiry or ticket, target zero.
- Alert handling: median time from high-severity alert to triage in the SIEM.
Hardening baseline
- Scope and tiers: decide which management groups get Defender CSPM and which workload plans, and record the cost model per resource type.
- Assign the benchmark: keep the Microsoft cloud security benchmark initiative assigned at the top management group; exemptions depend on it. See the Microsoft Cloud Security Benchmark Guide.
- Enable plans by policy: use Azure Policy to enable plans on every subscription, including new ones, and to deploy continuous export and workflow automation.
- Connect other clouds: onboard AWS and GCP at organisation level, review the roles created, and monitor them in CloudTrail and Cloud Audit Logs.
- Pilot remediation: start with one subscription, fix high-risk recommendations through owners, measure impact, and tune alert suppression rules for verified benign activity only.
- Route everything: continuous export to the SIEM, governance rules to owners, email notifications to a monitored mailbox.
- Lock down the control: Security Admin and Owner through PIM; Activity log alerts on plan disablement, export deletion and exemptions.
- Roll back safely: plans can be switched off per subscription (Plan 1 also per resource) if cost or performance needs review, but record it as an exemption with an expiry so coverage gaps stay visible.
Framework mapping
| Capability | NIST SP 800-53 Rev. 5 | NIST CSF 2.0 |
|---|---|---|
| Asset inventory and connectors | CM-8 | ID.AM-01 |
| Recommendations and benchmark assessment | CA-7, CM-6 | ID.RA-01 |
| Vulnerability and secrets scanning | RA-5 | ID.RA-01 |
| CIEM and role review | AC-6 | PR.AA-05 |
| Workload threat detection | SI-4 | DE.CM-09, DE.AE-02 |
| Export, automation, incident handling | IR-4 | RS.MA-01 |
Common misconfigurations
- Free tier only: relying on Foundational CSPM and believing attack paths and compliance are covered.
- Score chasing: fixing many low-risk recommendations for points while a single internet-exposed VM with a privileged identity stays open.
- Waivers without expiry: management group exemptions that silently cover every future resource.
- Resource Manager plan off: leaving the control plane unmonitored while protecting individual workloads.
- Connector roles edited or forgotten: widening AWS or GCP connector roles to fix an error, or leaving them in place after a connector is removed.
- Alerts to nowhere: notifications to subscription owners who left, and no export to the SIEM.
Practitioner checklist
- Decide Foundational versus Defender CSPM per management group, with the reason recorded.
- Enable Defender for Resource Manager and Key Vault on every subscription.
- Enable Servers, Storage, Databases and Containers plans where those workloads hold sensitive data.
- Move Security Admin and Owner to PIM-eligible assignments with approval.
- Inventory connector and auto-provisioning identities in Azure, AWS and GCP and monitor their activity.
- Configure continuous export to the SIEM and governance rules to resource owners.
- Require an expiry and ticket for every exemption and review them monthly.
- Alert on plan disablement, export deletion and MCSB initiative removal.
- Report risk-based score and attack path trends, not the classic score alone.
Standards and references
- Microsoft Learn: What is Microsoft Defender for Cloud?
- Microsoft Learn: Cloud security posture management (CSPM)
- Microsoft Learn: Select a Defender for Servers plan
- Microsoft Learn: Secure score in Defender for Cloud
- Microsoft Learn: User roles and permissions in Defender for Cloud
- Microsoft Learn: Permissions management (CIEM) in Defender for Cloud
- Microsoft Learn: Continuously export Defender for Cloud data
- MITRE ATT&CK: Disable or Modify Tools
Related NHI Mgmt Group resources: CNAPP Guide · Microsoft Azure Security Guide · Microsoft Sentinel Guide · Azure Policy and Management Groups Guide