TECHNOLOGY & CERTIFICATION EDITORIAL

What Is Microsoft Entra ID Conditional Access? Full Explanation

An administrator can configure a Conditional Access rule that looks sensible in the portal and still block the wrong people. An employee might be asked for multifactor authentication on every sign-in, a contractor might unexpectedly gain access from an unmanaged device, or an emergency administrator might be locked out while trying to repair the first two mistakes. The safest way to design Conditional Access is to treat it as an access decision you must be able to explain and test, not a collection of switches to enable.

Microsoft Entra ID Conditional Access decisions involve assignments, conditions, grant and session controls, report-only testing, sign-in evidence and emergency recovery. Before enforcing a rule, establish whose access should change, what result is expected for each test case, and which protected administrative path remains available if the rule blocks legitimate users.

Think in assignments, conditions and controls

A Conditional Access policy combines who it applies to, which resource is being accessed, optional signal-based conditions, and what the user must do or whether access should be blocked. Imagine a payroll application used by finance employees. A candidate rule might include a dedicated finance group, target that application and require multifactor authentication. Another might require a compliant device for access to sensitive administrative tools. The combination of assignments and controls determines the actual result; the policy name does not enforce anything by itself.

A useful design worksheet has five columns: included users or workload identities, excluded identities, target resources, relevant conditions and grant/session controls. Add an owner for the application and an explanation of the risk that the rule addresses. This catches the most common mismatch: a technically valid condition that is aimed at too broad a population or the wrong cloud resource.

Conditions may include aspects of the sign-in such as device state, location, client app and risk when supported and licensed. Do not assume every signal or policy type is available in every tenant. Confirm the applicable product requirements and licenses before promising that a specific risk condition can be enforced.

Define the audience before selecting MFA

Start with a pilot group that reflects the actual business role, not ‘all users’ simply because broad coverage sounds stronger. For a payroll example, include a small group of finance staff who use the application regularly, and separately identify administrative and guest scenarios that have different access needs. Target the payroll application rather than all cloud apps for the first test. Write down who should be included, who should be excluded, and what outcome the owner expects for each identity.

An exclusion is a safety-sensitive design choice. Emergency or break-glass accounts may need carefully governed exclusions so that an incident does not disable all administrative recovery paths. Those accounts should have dedicated protection and monitoring. Excluding an emergency account is not equivalent to leaving it unprotected or sharing its credentials. Validate the sign-in path and the organization’s recovery process before enforcing a policy that may affect administrators.

Do not combine user-oriented Conditional Access with the assumption that it automatically protects service principals. Workload identities and service principals have different access patterns and dedicated controls. A rule built around interactive users, browser sessions and MFA prompts may not cover unattended application-to-application authentication. Keep the intended identity type explicit in the design.

Understand grant versus session controls

Grant controls determine whether access is blocked or permitted after requirements are satisfied. A policy might block a sign-in outright, require a specific authentication strength, require multifactor authentication or require a compliant device when supported. When more than one grant control is selected, verify whether the configured combination requires all selected controls or allows one of them. Those are materially different security decisions. Do not infer the intended logic from a heading in the policy list.

Session controls govern aspects of the authenticated session after a sign-in meets the policy’s conditions. They are not a substitute for grant controls and can depend on app and platform support. For a sensitive browser application, you might need to limit a risky session differently from an ordinary compliant-device session, but the test must prove the app actually receives the intended restrictions.

Multiple applicable Conditional Access policies can affect one sign-in, and the result is not simply the outcome of the most recently edited rule. An identity might satisfy the payroll policy but be blocked by a separate organization-wide restriction. Always diagnose the full set of policies evaluated at sign-in instead of disabling the first policy that seems related to the complaint.

Walk through a finance-device scenario

Suppose the finance team requires MFA and a compliant managed device to open a payroll application. A reasonable initial proposal is:

Decision Pilot design Evidence to collect
Users and groups A small test finance group Included/excluded membership at sign-in
Target resource Payroll enterprise application Actual application/resource identifier
Grant controls Require MFA and compliant device, if supported Exact all-controls selection and result
Emergency path Documented, separately monitored emergency admin access Tested recovery sign-in and account protection
Initial state Report-only Expected outcomes in sign-in logs

Now test three accounts. A compliant finance user with an accepted MFA claim should meet the conditions. A finance user on an unmanaged device should reveal a device-related unmet requirement. An employee outside the targeted group should show that the finance-specific policy is not applied, though other policies may still apply. These are hypotheses to test—not guarantees about a particular tenant’s configuration.

A fourth case is essential: an administrator who needs to repair a failed application configuration. Verify that the emergency access procedure still works and that alerts or other independent controls make unusual emergency account usage visible. A policy cannot be called safely designed if it protects routine sign-ins but breaks recovery.

Report-only testing: what the result actually proves

Report-only mode evaluates most Conditional Access policies during sign-in and records the outcome without enforcing the grant and session controls of that report-only policy. It is therefore a simulation aid, not a partial production restriction. Microsoft documents some exceptions and user-experience caveats: User Actions scope is not evaluated in the same way, and compliant-device checks can still produce device-certificate selection prompts on certain mobile platforms even when enforcement is off.

In the sign-in logs, distinguish the following result categories:

Report-only result Practical interpretation
Success Applicable configured conditions and noninteractive requirements were satisfied
Failure Policy matched, but one or more relevant controls were not satisfied
User action required Matching policy would require an interactive step, such as an MFA prompt
Not applied The sign-in did not satisfy the policy’s conditions or assignments

Do not read ‘Report-only: Success’ as proof that the complete application or device is secure. It describes evaluation of that policy for that sign-in event. A healthy lab run includes both permitted and denied-or-unmet cases, with timestamps, affected account, resource, conditions and policy names. Keep copies of the expected and observed results; otherwise there is no evidence that the rule matched the right population.

Test a policy alongside existing enabled policies. A finance test user could already be subject to an organization-wide access restriction. Examine the Conditional Access and report-only result views together to establish whether the proposed payroll rule introduces the intended change or would duplicate existing enforcement.

Use sign-in logs and What If for separate questions

The What If tool is useful for modeling whether a proposed set of conditions and assignments would apply to a specified user, application and sign-in context. It does not reproduce every real-world client behavior or make an actual device compliant. Use it to refine scope, then inspect real controlled test sign-ins and their log entries for validation. A successful simulated match is only one piece of evidence.

Microsoft’s current What If documentation describes an evaluation API that needs representative sign-in inputs, including the actual application ID and conditions relevant to the policy. Missing a required location or risk signal can change a simulated policy match; a broad application grouping may not behave like a specific application ID. The tool also does not evaluate every service dependency—for example, a Teams scenario does not automatically model a related Exchange Online policy. Treat What If as a scope test, then confirm important decisions with appropriate real sign-in logs and authorized pilot cases.

When the pilot fails, start with its observable result rather than changing grant controls immediately. Check whether the policy applied, whether the account was included or excluded, whether the resource matches, which conditions were evaluated, which other policies applied and which control was unmet. If the user is asked for MFA repeatedly, investigate authentication claims, sign-in context and competing policies. If a noncompliant device is correctly denied, that may be intended behavior rather than a defect.

Treat policy troubleshooting like a small experiment: change one relevant assumption, retest with the same account and record how the outcome differs. Disabling every nearby policy can make the symptom disappear while leaving the actual control problem unidentified.

Roll out through monitored groups, not all at once

After the report-only evidence supports the intended audience and conditions, enable a carefully scoped pilot for the selected group. Confirm a real permitted sign-in and an intentionally unmet-policy case in an authorized test environment. Maintain a documented rollback owner who can disable or restore the policy through a protected administration path if unexpected disruption occurs.

Expand only after examining actual user behavior, support tickets and sign-in logs for the pilot. Consider guest accounts, service ownership, device-enrollment exceptions, shared workstations and noninteractive paths separately rather than treating them as ordinary employees. A policy that works for one compliant Windows laptop may need additional validation for supported macOS or mobile scenarios.

For high-impact apps, agree with the business owner what operational disruption is acceptable, who may authorize a scope increase and how affected users will regain access during an incident. The technical administrator cannot decide that alone. Preserve an as-built record of the policy’s assignments, controls, exceptions and reason for each deviation.

A troubleshooting map for common failures

Symptom Check first Avoid this shortcut
A permitted employee is blocked Included groups, targeted resource, other enabled policies and grant result Disabling every Conditional Access policy
Unexpected repeated MFA Sign-in context, previously satisfied claims and policy combinations Weakening MFA across the tenant
A pilot policy never applies Group membership, resource and conditions Assuming the policy is broken because of a misspelled name
Unmanaged device reaches payroll Whether compliant device was actually selected as a required grant, and how control logic is combined Assuming report-only testing enforced device restrictions
Administrator cannot recover Emergency access path and applicable policies Removing exclusions blindly or using shared accounts
User policy does not cover an automation Whether the caller is a service principal or workload identity Treating interactive-user CA as workload identity protection

Each row is a diagnostic starting point. Final decisions depend on verified tenant policy state, feature availability and logs. For example, a user outside the finance group may still be denied by another legitimate policy, and that does not make the finance policy incorrect.

Choose controls by the risk they mitigate

MFA addresses a different threat from device compliance. It can reduce the value of stolen passwords, but it does not prove a device is patched or managed. Device compliance establishes a defined device state, not that the user is trustworthy. Location checks can help in certain contexts but are an imperfect security signal and should not be treated as a substitute for identity and device assurance. Grant, block and session controls are most useful when you can explain the failure path each control is supposed to close.

For a payroll application, ask what matters: preventing password-only account compromise, preventing unmanaged-device data exposure, limiting the lifetime or capability of sessions, or constraining privileged operations. Then select testable controls with an owner and failure-recovery path. Adding more rules without mapping their interactions can make the system harder to understand and, paradoxically, harder to repair safely.

Turn the pilot results into an operational control

A policy should have a named owner, a recorded reason for each inclusion or exclusion, evidence from permitted and denied test cases, and a documented recovery procedure. Retain sign-in event identifiers and outcomes supporting the decision to leave report-only mode. Recheck the design whenever applications, authentication requirements or managed-device policies change: a safe rollout does not guarantee that every later sign-in will have the same result.

The final acceptance check is not ‘the policy says On.’ It is a documented set of expected sign-ins, observed results, an accountable owner and a working emergency path. If you cannot show how a legitimate user gets access, how a noncompliant case is handled, which policies contribute to that result and how to recover safely, the Conditional Access deployment is not yet complete.

Back to Insights
Explore what matters. Knowledge that goes beyond the exam.
Explore ExamTopics