Microsoft AZ-104: Azure RBAC for Platform Teams
Platform teams need a repeatable way to delegate Azure permissions without accumulating subscription-wide Owners. The design starts with service responsibilities, assignment scope, and evidence that temporary privilege is controlled. Platform teams often require broad visibility but narrowly scoped change rights across several subscriptions. Treat monitoring, resource deployment, identity administration, and security review as different tasks; inherited assignments and role definitions determine the actual blast radius.
On this page
- Translate platform responsibilities into permission boundaries
- Place scope where ownership is stable
- Use groups for humans and distinct identities for workloads
- Limit the authority to assign more authority
- Make privilege activation an ordinary controlled workflow
- Prefer built-in roles but examine what they truly allow
- Review the interaction with Policy, locks and service permissions
- Audit role assignments as changing infrastructure
Translate platform responsibilities into permission boundaries
A platform engineering group often operates shared networking, monitoring, subscription onboarding and deployment tooling. Those duties do not imply that every engineer needs permanent Owner rights to every application subscription. Assigning broad access for convenience makes it harder to separate a routine network change from an action that can grant new privileged roles. Role design in Microsoft AZ-104 should use the narrowest assignment scope that supports the required operation and then verify effective permissions.
Build a task inventory first. Which people maintain shared DNS? Which deploy a monitoring workspace? Who approves an application team’s request to access a production resource group? Who can change role assignments? Those questions identify different operations and resources. The role assignment should match a supported operation at an appropriate scope rather than encode the team’s seniority.
Azure RBAC applies a principal, role definition and scope. The role assignment model also supports descriptions and certain conditional authorization patterns. A reviewer needs to know why a permission exists and when it should end, not only that its technical assignment syntax is valid.
Place scope where ownership is stable
Management-group and subscription assignments can simplify access for a genuine central responsibility, but they also magnify mistakes. A network operations group responsible for a shared connectivity subscription may need durable authority there. It does not automatically need the same authority in every workload subscription that consumes the network.
Resource-group or resource scopes are often appropriate for application support teams. If a team owns an application VM fleet, granting access to a shared production identity service at subscription scope creates an avoidable exposure. The challenge is to choose boundaries that stay intelligible as resources and teams evolve; dozens of arbitrary one-resource exceptions may be difficult to maintain, while one high-level assignment may be too broad.
Inherited permissions should be reviewed. A user might have no direct assignment on a VM but still receive rights from the resource group, subscription or management group. A change review that inspects only the VM’s own IAM list may miss that inheritance. Make effective-access inspection a standard part of audits and support investigations.
Use groups for humans and distinct identities for workloads
Human permissions are easier to manage through purpose-specific groups than through unmanaged individual assignments across hundreds of resources. A group should have a clear owner and membership review, with separate groups for routine operations and privileged approval duties. The ease of using a group is not a reason to put every engineer into one all-powerful access group.
Automation has different lifecycle and attribution needs. A managed identity or service principal running a deployment should receive only the actions and scopes required by that pipeline. Do not use a human administrator’s credentials as a permanent deployment secret. Rotating one broad pipeline principal that serves many unrelated teams can be as disruptive as a human role design with no boundaries.
Records should distinguish human groups, workload identities and emergency access. A reviewer must be able to see why an application job has Blob Data Reader, why an engineer can restart VMs, and why only a limited set of privileged operators can modify those assignments. Reusing identities across unrelated jobs obscures that explanation.
Limit the authority to assign more authority
Owner, User Access Administrator and Role Based Access Control Administrator are especially sensitive because they can enable other role assignments within supported scopes. That authority is different from operating a resource. A person who can repair a VM need not necessarily assign another person permanent Contributor across a subscription.
Where supported, Azure role-assignment conditions can constrain delegated access management to specified roles or principals. Microsoft’s role assignment process documents options for conditional delegation. Review current support and test both permitted and forbidden assignments before declaring that a conditional delegation is safely restricted.
Separate the business approval from the technical act of assignment. A request should identify beneficiary, target scope, role, reason and duration. An access administrator can execute an approved change while a service owner remains accountable for whether the access is necessary. That division limits accidental permission expansion and creates useful audit evidence.
Make privilege activation an ordinary controlled workflow
Some operators need powerful rights rarely—during a subscription recovery, a governance rollout or a serious incident. Permanently active privileged roles increase exposure during the much longer periods when those powers are unnecessary. Microsoft Entra Privileged Identity Management can manage eligible and time-bound Azure resource roles where supported, with activation rules that may include approvals and other controls.
Use PIM deliberately. Define eligible scope, maximum activation duration, required justification and emergency exception handling. Test activation from the user’s point of view and review the associated audit trail. A platform team that must wait for an unavailable approver during a severe outage may need a governed break-glass process, not permanent high privilege for everyone.
A privilege system is only as strong as its recurring reviews. Remove obsolete eligibility when people move roles, and look for active permanent assignments that bypass the intended activation process. PIM does not automatically correct broad scope or an excessive role definition; it controls when the assigned authority becomes active.
Prefer built-in roles but examine what they truly allow
Built-in Azure roles reduce the need to maintain custom permission definitions. They are a good starting point, but their names are not sufficient security evidence. Check allowed management actions, any data actions, the ability to retrieve keys, and whether role assignment operations are included. Some broad management rights can indirectly enable data access through stored service keys.
If a built-in role is too broad, determine whether a narrower built-in role exists or a carefully reviewed custom role is justified. Custom roles increase maintenance responsibility as resource providers add APIs and actions. A role that lacks one needed action may lead teams to add wildcards; a role with overly broad actions may be riskier than its friendly name suggests.
Test the role using a representative principal and task. Confirm that it can perform the permitted operation and fails on a nearby unapproved one. This is more informative than seeing a role listed as assigned and assuming least privilege has been achieved.
Review the interaction with Policy, locks and service permissions
Azure RBAC is not the only reason an operation may fail. A deny Policy effect can reject noncompliant configuration, a resource lock can restrict management changes, and a data-plane operation may require a role different from VM administration. Platform support should use error evidence to find the controlling layer rather than repeatedly escalating the user’s role.
Consider a production storage account that an engineer can manage but cannot read through Microsoft Entra data authorization. That separation may be intentional. Giving subscription Owner as a “fix” could introduce a broader key-access path without resolving the intended data-role model. Azure RBAC scope must be interpreted alongside the service’s data-plane rules and current authorization guidance.
Document exception processes separately for each control plane. The team that approves Policy exemptions may not be the same one responsible for emergency identity elevation. Keeping the responsibilities clear helps operators make an authorized correction under pressure.
Audit role assignments as changing infrastructure
A role graph evolves with subscriptions, contractors, service principals and deployment pipelines. Periodically export or inspect relevant assignments, group membership and privileged eligibility, then reconcile them against the task inventory. Broad unowned assignments and long-expired emergency exceptions deserve attention even if no incident has occurred.
Record why each important assignment exists, its expected scope, owning team, and retirement condition. Audit logging and access reviews should also establish whether the right person can revoke privilege quickly after a compromise. A security control that requires several undocumented manual steps to remove an account is weaker than it appears on a policy slide.
In a disposable resource group, test delegated access with two identities representing different responsibilities. Give the workload operator the minimum role necessary to read health and perform approved updates, while requiring separate authorization for changing role assignments. Demonstrate both permitted operations and denied privilege escalation; then inspect whether a group membership or parent-scope assignment unexpectedly broadens access. Record role definition, principal, scope, effective permission and the process for time-limited exceptions. The production goal is a platform team that can support services without inheriting unneeded control over every subscription resource.