When you connect an Azure subscription to CloudMonitor, our scanner does one job: read the Azure Resource Manager (ARM) API as a read-only observer and turn what it sees into findings. This post lists the scanners that run against an Azure subscription today, and what each one flags.
1. Compute
- Virtual machines (utilization) — power state plus seven days of CPU metrics from Azure Monitor, classified so idle and oversized VMs stand out.
- VM, scale-set and disk security — a public IP attached directly to a VM NIC, a network security group that allows broad inbound access, scale sets with public IP configuration or no autoscale setting, and managed disks attached to nothing.
- OS and software end-of-life — VM images compared against the public endoflife.date catalog, flagged when past or approaching end of life.
- App Service and Function Apps — minimum TLS version, FTP left enabled, end-of-life runtimes, expired or expiring certificates, and no managed identity.
- Container Apps and Container Registry — external ingress without platform authentication, registry admin user enabled, and similar configuration findings.
- AKS clusters — Kubernetes version and available upgrades, node-pool status and node image age, RBAC and Entra integration, network plugin. See the Kubernetes post.
2. Storage & data
- Storage accounts — publicly accessible containers, public network access, encryption, blob soft-delete, access keys that have not been rotated, plus the waste that tends to gather around them: unattached disks, old snapshots, orphaned NICs and unused public IPs.
- Azure SQL Database, PostgreSQL and MySQL Flexible Server — public network access, minimum TLS, backup retention under seven days, transparent data encryption (Azure SQL), Entra-only authentication, geo-redundant backup. Azure SQL Managed Instance is not covered.
- Cosmos DB — public network access and firewall scope, key rotation age (from timestamps only — we never list keys), multi-region and failover configuration, backup mode.
- Key Vault — every secret, key and certificate by name and expiry date, never by value. Expiry alerts follow one ladder: 90, 60, 30, 14, 7 and 1 day before expiry, and again once it has expired, with a recovery notice once the item is renewed. The vault itself is checked too: soft delete off or with short retention, purge protection off, public network access with no firewall, and vaults still on the legacy access-policy permission model.
- Backups — Recovery Services and Backup vaults (soft delete, immutability, storage redundancy, failed jobs in the last seven days, when a restore was last run), each protected item's last backup against its policy, VMs and storage accounts that no vault protects, and the built-in point-in-time retention and geo-redundant backup of your Azure SQL, PostgreSQL and MySQL databases.
3. Networking
Network security groups are the high-signal target: an inbound Allow rule open to the whole internet on a sensitive port — SSH, RDP, SQL Server, MySQL, PostgreSQL, MongoDB, Redis, Elasticsearch and similar — or on every port. We also read public IP addresses (flagging ones attached to nothing), load balancers (rules with no health probe) and Application Gateways (WAF disabled, HTTP listeners with no HTTPS redirect, expired or expiring listener certificates). In public Azure DNS zones, a CNAME that points at a resource which no longer exists is flagged as a possible subdomain takeover.
4. Changes, Microsoft notices and governance
- Activity Log — who changed what. Administrative writes, deletes and actions are recorded, and the risky ones are flagged: role assignments and definitions, elevated access, policy removals and exemptions, lock removal, NSG and firewall rule changes, Key Vault access changes and deletion, deleted diagnostic settings, federated credentials, VM run-command and disk export. We keep who, what and when; we never store the request body, which for some operations carries secrets. The first scan reads back seven days so the timeline is not empty, without emailing you about that history. Reader is enough.
- Service Health and Advisor — active Azure incidents in the regions your subscription uses, and retirement notices for features your resources depend on, each with its retirement date.
- Governance — production resource groups and critical resources with no lock, key resources with no diagnostic setting sending logs anywhere, non-compliant Azure Policy assignments, resource groups missing required tags, and Defender for Cloud plans left on the free tier.
- Quota headroom — usage against vCPU, network, storage and Container Apps quotas in the regions you use, App Service plans close to their instance ceiling, and Key Vault throttling, flagged at 80% and 95%.
5. Scanners that need an extra role
Three scanners run against every Azure connection but only find something once you grant more than Reader. Without the extra grant they skip quietly and the rest of the scan carries on:
- Identity (Microsoft Entra ID, through read-only Microsoft Graph application permissions that a directory admin grants) — orphaned and guest accounts, MFA gaps, standing privileged roles, app registrations with expiring secrets or certificates, and the tenant's Conditional Access and admin-role posture against the CIS Microsoft 365 benchmark, app consent risk (who can consent to apps, apps holding high-risk permissions, apps with no owner or no recent sign-in), and Microsoft 365 license waste with an estimated monthly cost at list price. Each permission is independent: leaving one out switches off only the findings beside it, the scan still runs, and the subscription's status names the one that is missing.
- Directory.Read.All — guest accounts and accounts whose password never expires, plus the user list behind the inactive-account check; delegated permissions consented to apps; Microsoft 365 license counts and each user's licenses; and each Microsoft 365 subscription's status and renewal date.
- AuditLog.Read.All — inactive accounts (last sign-in), users with no MFA method registered, administrators with weak MFA methods, subscription Owners with no MFA, apps and licensed users with no recent sign-in, and PIM-eligible admin roles nobody has activated (P2). Sign-in and MFA data need a Microsoft Entra ID P1 or P2 license, and Microsoft asks for this permission to be granted together with Directory.Read.All.
- RoleManagement.Read.Directory — privileged role assignments such as Global Administrator, whether each is permanent or eligible in Privileged Identity Management (P2), and the Global Administrator count. Directory.Read.All also covers the assignment list.
- Application.Read.All — app registrations with expiring or expired client secrets and certificates, the active SAML signing certificate of each enterprise application, the permissions each app holds, and app registrations with no owner. Directory.Read.All also covers this read.
- Policy.Read.All — security defaults and Conditional Access policies: MFA for all users and for administrators, legacy authentication, risk-based policies (P2), break-glass exclusions and policies left in report-only mode; whether users can consent to or register applications.
- DeviceManagementServiceConfig.Read.All — only for tenants that use Microsoft Intune with Apple devices: when the Apple MDM push certificate, volume-purchase (VPP) tokens and Automated Device Enrollment tokens expire. Expiry dates only; a tenant without Intune is never asked for it.
- SecurityEvents.Read.All — your Microsoft 365 Secure Score: the current score and the maximum, once a day, for the trend and the change over 30 days. Only those two numbers are read and stored.
- ServiceHealth.Read.All — Microsoft 365 Service Health: the incidents and advisories Microsoft currently reports for the services your tenant uses, checked every 5 minutes on a paid plan (every 15 during a trial). Service status only; no mailbox, message or user data.
- ServiceMessage.Read.All — Microsoft 365 Message Center: announcements of feature retirements and major changes, counted down to their action-required date. Titles, services and dates only.
You read the results in the product under Findings → Azure → Directory, grouped by what was found and filterable by severity. Before the grant is in place that tab says so, and names what it is still waiting for.
- Reservations and contracts (a reservations or billing reader role) — reserved-instance expiry, alongside contract and license dates you enter by hand.
- Defender for Cloud (Security Reader) — your secure score and the recommendation types currently unhealthy, read every 5 minutes on a paid plan (every 15 during a trial). A summary, not a re-implementation of Defender.
6. Cost and waste
Every night a spend check reads each subscription's daily usage, flags a day whose spend jumps well above the median of the week before it (or a new service that suddenly costs money), and reads your Azure budgets to flag spend reaching 75, 90 and 100% of a budget, or Azure's own forecast heading past it. That needs only Reader, and we never create a budget for you. A waste scan prices what costs money while doing nothing — unattached disks, old snapshots, public IPs and network interfaces attached to nothing, empty paid App Service plans, idle load balancers and NAT gateways — at Azure list price, and totals it per subscription. The Cost Explorer pages read the Azure Cost Management API (it needs Cost Management Reader) and cache spend by service; an account admin refreshes them from the page.
What the Azure connector does NOT do
- It does not read data-plane contents. We never enumerate the blobs inside a storage account, the rows inside a database, the secret values inside Key Vault (we only read names, expiry, and metadata), or the messages in a queue.
- It does not write. The roles we ask for are Reader, Key Vault Reader and Monitoring Reader, plus the optional read-only grants above. CloudMonitor makes no write call to your cloud — where we suggest a fix, we show you the command and you run it.
- It does not open a tunnel into your network. Everything is pull-mode against public ARM endpoints. There is no agent.
- It does not store what a change contained. The Activity Log scan records who changed what and when, never the request body.
How often it runs
It depends what the scan is looking for. Findings Azure has already raised — Defender for Cloud recommendations, risky Activity Log changes, Service Health incidents — are read every 5 minutes on a paid plan (every 15 during a trial). Exposure and state — storage and database public access, NSG rules open to the internet, public IPs, unauthenticated container ingress, App Service and AKS state, registry admin users — is re-checked every hour, because any of it can change the moment someone clicks something. Posture and expiry — Key Vault secrets and certificates, backups, governance, quotas, DNS, end-of-life dates, MuleSoft runtime versions — runs once a night, in a single pass, because its warning ladder steps in days. Utilization, cost and waste ride that same nightly pass. Endpoint and certificate monitoring runs every 5 minutes on a paid plan (every 15 during a trial). The cadence is the same for every customer; there is no setting or tier that changes it. An account administrator can ask for a fresh scan of a connection, at most once every 15 minutes per account; it runs with the next hourly scan.
The read-only roles we ask for are on the Connectors page.