AB-650 Administering Microsoft 365 and AI Services Exam Topics and Questions
These Microsoft Administering Microsoft 365 and AI Services (AB-650) exam topics are organized according to official exam domains to help candidates quickly verify coverage and focus on assessment rather than theory. Each domain is paired with topic-wise AB-650 sample questions that reflect how objectives are tested in the actual exam. This structure enables efficient review, targeted self-assessment, and rapid identification of weak areas when preparing for the Microsoft Administering Microsoft 365 and AI Services certification exam.
Let's Practice Free Microsoft AB-650 Questions Aligned with Official Exam Topics
This topic is the operational foundation of the exam. Before any governance or AI feature can work correctly, the tenant itself has to be configured right: domains resolved, licenses assigned, services healthy, and workloads in a state where users can actually do their jobs. Errors here ripple outward into every other area, so examiners treat it as a proving ground for breadth. Expect to be held accountable for specifics, not general familiarity. – Tenant setup and organisational configuration Tenant configuration...
This topic is the operational foundation of the exam. Before any governance or AI feature can work correctly, the tenant itself has to be configured right: domains resolved, licenses assigned, services healthy, and workloads in a state where users can actually do their jobs. Errors here ripple outward into every other area, so examiners treat it as a proving ground for breadth. Expect to be held accountable for specifics, not general familiarity.
– Tenant setup and organisational configuration
Tenant configuration covers the decisions made once and rarely revisited, but which shape everything else. Branding elements such as company logo, sign-in backgrounds, and custom themes are configured per tenant and sometimes per group; knowing which settings live in the Microsoft 365 admin center versus Entra is testable. Domain implementation involves adding, verifying, and setting a domain as the default, and the DNS record types required for each service (MX, CNAME, TXT) are fair game. Organisational settings span security and privacy defaults, the organisation profile, and which self-service options are exposed to users. Network connectivity insights sit in the admin center and surface per-location metrics; the exam tests whether you know what those reports tell you and what actions they recommend, not just that the feature exists. Microsoft 365 Backup adds setup, restore scope, and monitoring responsibilities that are distinct from third-party backup workflows.
– Licensing, including AI service licences
Licence management is tested at a more granular level than most candidates expect. Group-based licensing means understanding assignment conflicts, processing errors surfaced in Entra, and what happens when a user is removed from a licensing group. Pay-as-you-go billing for services like Teams Premium or Communication Credits introduces a separate billing model that sits alongside subscription licences. AI service licences add another layer: Microsoft 365 Copilot, Microsoft Agent 365, and Copilot Studio each have their own licence types, and the exam distinguishes between assigning a Copilot licence and enabling the features that licence unlocks. Monitoring licence consumption, identifying unassigned licences, and reacting to licence shortfalls are all practical tasks the exam will probe. Knowing which admin center surface shows which licence data saves time in scenario questions.
– Service health and monitoring
Service health monitoring is a narrow but reliable area. The Service health dashboard in the Microsoft 365 admin center shows advisories and incidents; the exam tests what each status means, how to configure email and webhook notifications for health events, and how to find the history of a past incident. Candidates who treat this as a read-only dashboard miss the notification configuration piece, which is where items are set. Network connectivity insights, covered under tenant setup above, feed into this operational picture but are distinct from service health. The distinction between an advisory (degraded performance) and an incident (service outage) matters because the recommended admin response differs between them.
– Workload administration: Exchange, Teams, and SharePoint
The workload section covers three services and expects practical depth in each. For Exchange, shared mailboxes are the most tested object type: creation, permission delegation (Full Access versus Send As versus Send on Behalf), and the licensing rules that apply. For Teams, the exam tests channel types (standard, private, shared), ownership rules, and membership management, plus the Copilot-specific settings for meetings, including whether transcription is required before Copilot features activate. SharePoint site creation, permission inheritance, and the sharing model are foundational, but the Copilot-specific layer adds SharePoint Advanced Management, Microsoft Search configuration, and site exclusions from Copilot's index. OneDrive for Copilot follows the same pattern: the admin controls that determine which content Copilot can reach are examinable alongside the core OneDrive settings. Candidates who study Exchange, Teams, and SharePoint in isolation without connecting them to Copilot readiness will find gaps in scenario questions.
How Configure and manage Microsoft 365 tenants and workloads is tested
This topic carries the largest share of the exam, and the items reflect that breadth. Scenario questions place you in the role of an admin who has received a specific complaint or task, then ask which admin center, which setting, or which sequence of steps resolves it. The trap is selecting a plausible-sounding action that touches the right service but the wrong setting. For example, items about shared mailboxes reward knowing the exact permission model rather than the general concept of delegation. Items about Copilot in Teams meetings reward knowing the dependency on transcription being enabled before Copilot features are available. Licence-related items reward knowing where processing errors surface and what they mean, not just that group-based licensing exists. Service health items reward distinguishing between incident types and knowing which notification mechanism applies to each. Candidates lose marks by picking the correct service but the wrong configuration surface, or by applying a general principle where the exam expects a specific administrative step.
A large question bank lets you see exactly which sub-areas within this topic keep catching you out. Running the practice test more than once across the full topic reveals whether your misses are clustered around licensing, workload-specific settings, or Copilot configuration, so you can target your revision precisely rather than re-reading everything.
The question below tests one of the workload configuration scenarios from this topic. Work through it before checking your reasoning.
You have a Microsoft 365 E5 subscription that uses Microsoft Defender for Office 365. Strict protection is enabled for all recipients.
The subscription contains a group named Group1 that contains all the mailboxes for your company's marketing department.
Several marketing department mailboxes recently sent high volumes of external email after a credential was compromised.
You need to enforce a lower external message limit for Group1. The solution must automatically restrict senders that reach the limit.
What should you do?
Where the first topic is about getting services running, this one is about controlling who can reach them and what they can do. Identity management, authentication hardening, threat response, and data protection form four distinct disciplines here, and the exam tests all of them at a level that requires knowing the right tool for a given scenario. Picking the wrong Entra feature or the wrong Purview policy type is the most common way to lose marks in this area. –...
Where the first topic is about getting services running, this one is about controlling who can reach them and what they can do. Identity management, authentication hardening, threat response, and data protection form four distinct disciplines here, and the exam tests all of them at a level that requires knowing the right tool for a given scenario. Picking the wrong Entra feature or the wrong Purview policy type is the most common way to lose marks in this area.
– Identity and access administration in Microsoft Entra
User and group management in Entra is tested at the level of object properties and lifecycle, not just creation. Guest users introduce external collaboration settings, and the exam distinguishes between Entra B2B invitation policies and cross-tenant access settings. Administrative units let you scope admin roles to a subset of users or devices; the exam tests when to use them and how role assignments interact with them. Privileged Identity Management adds just-in-time role activation, approval workflows, and access reviews to the standard role assignment model. Candidates who know only permanent role assignment will miss PIM-specific scenarios. Bulk management via Microsoft Graph PowerShell is tested practically: knowing which cmdlets handle which object types, and when PowerShell is the right tool over the admin center, is expected. Contacts are a smaller sub-area but appear in scenarios involving external recipients in Exchange.
– Authentication and Conditional Access
Authentication configuration covers the methods available in Entra (FIDO2, authenticator app, SMS, OATH tokens, certificate-based auth), how to enable or restrict them at the tenant and per-user level, and how to read the authentication methods activity report. Entra Password Protection enforces banned password lists, both the Microsoft global list and custom lists, and can be applied on-premises through the Entra Password Protection agent. Self-service password reset requires knowing which authentication methods are valid for reset, how to configure the registration campaign, and what the writeback option requires. Conditional Access policies are the most scenario-heavy sub-area: the exam tests policy conditions (user, location, device compliance, sign-in risk), grant controls (MFA, compliant device, hybrid join), and session controls. Entra ID Protection feeds risk signals into Conditional Access; knowing which risk levels trigger which policy responses is directly testable. Investigating authentication issues means reading sign-in logs and identifying the specific failure reason.
– Threat protection with Defender for Office 365
Defender for Office 365 is tested through its operational workflow rather than its marketing tiers. Alert management means knowing where alerts surface, how to triage them by severity and category, and when to escalate versus remediate directly. Threat policies cover anti-phishing, anti-spam, anti-malware, Safe Links, and Safe Attachments; the exam tests which policy type addresses which threat vector and what the configuration options within each policy control. Investigating email threats involves the Threat Explorer and the Email Entity page: knowing what each view shows and how to trace a message through the mail flow is expected. Attack simulation training is tested at the configuration level: choosing a simulation technique, assigning training based on simulation results, and reading the campaign report. Candidates who know Defender conceptually but have not worked through the policy configuration options will lose marks on specificity.
– Data protection with Microsoft Purview
The Purview sub-area is framed around identifying requirements rather than deep technical configuration, which shifts the exam demand toward decision-making. For DLP, the exam tests which workload a policy applies to, what conditions and actions are available per workload, and how Copilot interactions are treated as a policy scope. Sensitive information types and sensitivity labels underpin both DLP and information protection; the exam distinguishes between a label, a label policy, and the conditions that trigger auto-labelling. Retention labels and retention policies are separate objects with different precedence rules, and the exam tests what happens when both apply to the same content. DLP alerts for AI services introduce a newer scope: Copilot interactions can trigger DLP policy matches, and reviewing those alerts follows a similar workflow to standard DLP alerts but requires knowing the Copilot-specific policy conditions. Purview DSPM for AI adds a posture layer: it surfaces risky AI activity and data exposure signals rather than enforcing policy directly.
How Govern and secure Microsoft 365 tenants and workloads is tested
Items in this topic reward choosing the right tool from a set of plausible alternatives. A scenario might describe a requirement that could be addressed by Conditional Access, Identity Protection, or PIM, and the item rewards the one that fits the specific constraint in the scenario. The same pattern appears in Purview: a data protection requirement might fit a DLP policy, a sensitivity label policy, or a retention policy, and the item rewards matching the requirement to the correct object type. Authentication items reward knowing the specific failure reason in a sign-in log rather than a general troubleshooting approach. Defender for Office 365 items reward operational knowledge: which console surface shows which data, and which action resolves which alert type. Candidates lose marks by picking a feature that is conceptually related but scoped incorrectly, or by confusing Entra objects that sound similar. Reading the scenario conditions carefully before selecting is the discipline this topic demands.
Checking a weak area in this topic is more efficient under timed conditions than re-reading documentation. The practice test lets you work through identity, authentication, Defender, and Purview scenarios separately, so you can confirm which sub-area is solid and which needs another pass before exam day.
The question below presents a governance or security scenario from this topic. Commit to an answer before reviewing your thinking.
You have a Microsoft 365 subscription that uses Microsoft Defender for Office 365.
You need to ensure that users are prevented from opening or downloading malicious files from Microsoft Teams, OneDrive, or SharePoint Online.
What should you do?
This topic covers the layer of the Microsoft 365 platform that most organisations are still rolling out, which means exam questions here are built on official documentation rather than years of community practice. The sub-topics split into three areas: enabling and configuring Copilot for users, managing agents and the agent registry, and monitoring AI services across cost, usage, and health. Getting the sequencing right matters: readiness and data governance decisions come before deployment, not after. – Copilot enablement and tenant...
This topic covers the layer of the Microsoft 365 platform that most organisations are still rolling out, which means exam questions here are built on official documentation rather than years of community practice. The sub-topics split into three areas: enabling and configuring Copilot for users, managing agents and the agent registry, and monitoring AI services across cost, usage, and health. Getting the sequencing right matters: readiness and data governance decisions come before deployment, not after.
– Copilot enablement and tenant configuration
Tenant readiness for Copilot is assessed across several dimensions before a licence is useful: data governance, oversharing controls, and in-app experience settings all need to be in the right state. The exam tests how to identify oversharing risks (files with broad permissions that Copilot could surface to the wrong users) and what tools address them, including SharePoint Advanced Management controls covered in topic one. Web search for Copilot and Microsoft 365 Copilot Chat is a toggle with policy implications: knowing which admin surface controls it and what the off state means for Copilot responses is testable. Copilot Search configuration is a separate setting from general Microsoft Search. Tenant-level Copilot settings include self-service purchase controls, whether Copilot appears in admin centers, release preference channels, AI-generated content disclaimers, and video and image generation permissions. The Cowork feature and third-party AI provider management are smaller sub-areas but appear in scenarios about controlling which AI capabilities users can access.
– Agent management and the agent registry
The agent registry is the central catalogue for Microsoft-built, third-party, and custom agents available in the tenant. The exam tests the full lifecycle: discovering agents in the registry, reviewing submissions, and publishing or rejecting them. Installing, blocking, or restricting agent access follows a similar model to Teams app management but is distinct from it. Uploading custom agents has its own set of requirements. Agent settings cover which agent types are permitted, sharing permissions, template availability, and user access scope. Agent owners are a defined role with specific responsibilities; the exam distinguishes owner actions from admin actions. Copilot connectors extend the data sources available to Copilot and agents; configuring them involves authentication and data scope decisions. Agent 365 introduces tools management, and the exam tests which tools can be enabled or restricted within Agent 365 and what the administrative controls look like.
– Agent identity, access, and security
Agents in Microsoft 365 have identities managed through Microsoft Entra Agent ID, which means the lifecycle workflows for agent identities (creation, modification, decommissioning) follow Entra patterns but apply specifically to non-human principals. The exam tests what Entra Agent ID manages and how it differs from managing a service principal or managed identity in general Entra contexts. Securing agent access uses access packages from Entra Identity Governance and Conditional Access policies; the exam tests how these controls apply to agent principals rather than human users. The distinction matters because some Conditional Access conditions behave differently for workload identities. Agent 365 adds a governance and security layer: protecting sensitive data that agents can reach, evaluating compliance gaps in agent behaviour, and monitoring agent activity all have dedicated capabilities in Agent 365 that are tested at the feature-and-action level.
– Monitoring costs, usage, and service health for AI services
AI service monitoring is tested across three dimensions. Cost management for Copilot and agent services involves tracking consumption against licence allocations and pay-as-you-go meters; the exam tests where these figures are visible and what actions are available when costs exceed expectations. Usage and adoption reporting is split between the Microsoft 365 admin center (workload-level Copilot adoption details) and the Copilot Control System, which provides a broader view of AI service usage across the tenant. The Copilot Control System also surfaces service health signals specific to AI services, separate from the main Service health dashboard. Candidates need to know which report lives in which surface and what each report shows, because scenario questions will describe a monitoring requirement and expect you to name the correct tool. Conflating the admin center usage reports with Copilot Control System data is a reliable way to lose marks in this sub-area.
How Manage and secure AI services in Microsoft 365 is tested
Items in this topic reward knowing the administrative sequence and the correct console surface for each AI management task. Because these features are newer, the exam leans on scenario-based items that describe a business requirement or a problem state and ask which configuration step resolves it. Copilot readiness items reward identifying the specific gap (oversharing, missing licence, wrong release channel) rather than a general readiness review. Agent registry items reward knowing the difference between discovering, publishing, installing, and blocking an agent, because those are four distinct actions with different outcomes. Agent identity items reward applying Entra concepts to non-human principals, where the instinct to use human-user workflows is the trap. Monitoring items reward matching a reporting requirement to the exact surface that shows that data. Candidates lose marks by selecting a monitoring tool that is adjacent to but not the one that covers AI-specific signals, or by applying a standard Entra identity workflow to an agent identity scenario where Agent ID is the correct scope.
Because this topic covers features that change frequently, seeing the item style before the exam matters as much as knowing the content. The free demo of the practice test gives you a sample of how these AI service scenarios are framed, so you know what the question format demands before you sit the real thing.
The question below targets one of the AI service management or monitoring scenarios from this topic. Read the scenario carefully before selecting.
Ready to Start Practicing?
Access all questions and start your exam preparation journey
Upgrade to Full AB-650 Exam Questions 🚀
