Domain 1: Identify the Core Features and Objects of Microsoft 365 Services
- Licences decide what a user can reach. A licence assigned directly to a user and one inherited from a group both grant access, and group-based licensing is how organisations scale it - but a user without the Microsoft 365 Copilot licence sees no Copilot regardless of any other setting.
- The Microsoft 365 admin center is the top-level console: domain names, organisation settings, users and licences, service health, and the links out to the specialist admin centers. Anything described as an org-wide setting is usually configured here.
- The Exchange admin center manages mail objects - mailboxes, shared mailboxes, distribution groups and mail flow rules. A distribution group delivers mail to its members; a Microsoft 365 group additionally carries a shared mailbox, calendar and SharePoint site.
- The SharePoint admin center manages sites, and within a site the content lives in libraries and folders. Site-level access comes from the SharePoint groups - Owners with full control, Members who can edit, Visitors who can read - and permissions inherit down to libraries and items unless inheritance is broken.
- The Teams admin center manages teams, channels and the policies that control what users may do, such as meeting, messaging and app permission policies. Every team is backed by a Microsoft 365 group and a SharePoint site, which is why Teams files are really SharePoint files.
- Know which console owns which object, because a large share of the exam is exactly that mapping: users and licences in the Microsoft 365 admin center, mailboxes in Exchange, sites and libraries in SharePoint, teams and channels in Teams, identity in Microsoft Entra, and compliance in Microsoft Purview.
- Zero Trust rests on three principles: verify explicitly using all available signals, use least-privilege access with just-in-time and just-enough rights, and assume breach by segmenting and verifying end to end. The network being internal is not evidence of trustworthiness.
- Authentication proves who a user is; authorisation decides what that proven identity may do. Conditional access sits between them, evaluating signals after authentication succeeds to decide whether access is granted, blocked or challenged.
- Authentication methods differ in strength: passwordless options such as Windows Hello for Business, FIDO2 security keys and passkeys are strongest, the Microsoft Authenticator app with number matching is strong, and SMS is the weakest of the accepted second factors.
- Microsoft Defender XDR correlates signals across identities, endpoints, email and cloud apps into a single incident rather than a pile of alerts, which is what lets an analyst see one attack instead of five unrelated warnings.
- Microsoft Entra ID is the identity platform underneath everything: it holds the users and groups, issues the tokens, and enforces the conditional access policies. Every Microsoft 365 sign-in goes through it.
- Conditional access policies are if-then rules: for these users and apps, under these conditions - device state, location, sign-in risk - require multifactor authentication, require a compliant device, or block. Always exclude a break-glass account so a policy cannot lock everyone out.
- Single sign-on lets a user authenticate once and reach many applications without signing in again, which improves security as well as convenience: fewer prompts means fewer passwords and one place to enforce policy and revoke access.
- Choose the right security object: assign permissions to groups rather than individual users so membership changes handle joiners and leavers, and use a security group for permissions and a Microsoft 365 group when the members also need shared collaboration resources.
- Troubleshoot sign-in problems with the Microsoft Entra sign-in logs, which show the exact failure reason - which conditional access policy applied and whether it granted or blocked, whether MFA was satisfied, and whether the sign-in was flagged as risky.
- Identity Secure Score scores your identity configuration against Microsoft recommendations as a percentage with specific improvement actions. It measures configuration, not whether you have been attacked, and it is a prioritisation tool rather than a compliance certificate.
- Audit logs record user and admin activity across the services and are searched in Microsoft Purview, which is where you answer who deleted a file, who changed a policy, or who accessed a mailbox. Retention of those logs depends on the licence.
- Privileged Identity Management makes administrative roles eligible rather than permanent: an admin activates the role when needed, for a limited time, optionally with approval and justification. That removes standing privilege, which is the point.
- Distinguish app registrations from enterprise applications: a registration defines an application and its permissions in your tenant, while an enterprise application is the instance of an application - yours or a third party's - that users actually consent to and sign in to.
Domain 2: Understand Data Protection and Governance Tasks for Microsoft 365 and Copilot
- Microsoft Purview is the compliance and data governance suite, and this domain is the largest on the exam. Learn what each component is for, because most questions are a scenario matched to the right tool rather than a configuration detail.
- Purview Information Protection classifies and labels content, and a sensitivity label can enforce encryption, watermarking and access restrictions that travel with the file wherever it goes - including out of the organisation.
- Data Loss Prevention detects sensitive content in transit and acts on it: block the send, warn the user with a policy tip, or allow with an override and justification. DLP policies apply across Exchange, SharePoint, OneDrive, Teams, endpoints and now AI prompts.
- Insider Risk Management looks for risky behaviour by people inside the organisation - mass downloads before a resignation, exfiltration to personal storage - using indicators and policy templates, with the investigation pseudonymised by default.
- Communication Compliance scans messages in Teams, Exchange and other channels for policy violations such as harassment, offensive language or regulatory breaches, and routes matches to reviewers.
- Data Security Posture Management for AI is the Copilot-era addition: it discovers AI activity across the organisation, shows what data is being fed to AI tools, flags risky prompts and unlabelled sensitive data reaching AI, and recommends policies. Expect it as the answer to "how do we see what AI is doing with our data".
- Data Lifecycle Management governs how long content is kept and what happens after: retention labels and policies keep content for a defined period and then delete it or start a disposition review.
- Sensitivity labels and retention labels solve different problems and are often confused. Sensitivity labels control who can open content and what they can do with it; retention labels control how long it is kept and whether it can be deleted.
- Data classification identifies sensitive information using sensitive information types, trainable classifiers and exact data match, which is what DLP and auto-labelling policies act on. Classification finds the data; labels and policies decide what happens to it.
- Retention resolves conflicts by fixed rules: retention always wins over deletion, the longest retention period wins, an explicit label beats an inherited policy, and a shorter deletion loses. Those four rules answer most retention questions.
- Copilot accesses only data the signed-in user can already access. It does not bypass permissions, so it cannot show a user a document they could not open themselves - which means Copilot exposes existing oversharing rather than creating new access.
- Microsoft Graph is what grounds Copilot in your organisation's content: it retrieves relevant files, emails, chats and meetings the user has access to, and those become the context for the response. Better-organised, correctly permissioned content produces better answers.
- Copilot inherits the existing controls: Entra ID for identity and conditional access, sensitivity labels that are honoured and carried through to generated content, DLP policies that apply to prompts and responses, and Defender for threat protection.
- Prompts and responses are not used to train the foundation models, and organisational data stays within the Microsoft 365 service boundary and its compliance commitments. That is the standard answer to a data-residency or model-training question.
- Microsoft's responsible AI principles are fairness, reliability and safety, privacy and security, inclusiveness, transparency, and accountability. Expect to match a described concern to the right principle rather than to list them.
- Compliance Manager scores your compliance posture against regulations and standards and gives specific improvement actions with points, splitting responsibility between what Microsoft manages and what you must do.
- Use the right Purview tool for the question being asked: Data Explorer to see the sensitive content that exists, Activity Explorer to see what users did with labelled content, DSPM for AI to see AI activity specifically, and Content search in eDiscovery to find specific files and emails.
- Respond to a DLP alert in the Purview alerts view: see which policy matched, what content triggered it, who was involved and whether it was blocked or overridden, then tune the policy if it is producing false positives rather than switching it off.
- Oversharing is the central Copilot governance risk, because Copilot surfaces everything a user may access. A site shared with Everyone except external users has always been overshared - Copilot just makes it visible.
- Run a data access governance report in the SharePoint admin center to find sites shared with everyone, sites with anonymous links, and sites with the most sharing activity. It is the starting point for reducing oversharing before a Copilot rollout.
- SharePoint Advanced Management adds the controls to act on those findings: restricted access control limits a site to a specified group regardless of individual permissions, restricted content discovery keeps a site out of Copilot and search results, and site lifecycle policies clean up inactive sites.
Domain 3: Perform Basic Administrative Tasks for Copilot and Agents
- Distinguish Copilot from agents: Microsoft 365 Copilot is the built-in assistant across Word, Excel, PowerPoint, Outlook and Teams, grounded in your Microsoft Graph data, while an agent is something you create for a specific purpose with its own instructions, knowledge and actions.
- Compare the two licensing models: the Microsoft 365 Copilot licence is a fixed monthly per-user charge covering the built-in experiences, while pay-as-you-go bills consumption against an Azure subscription - which is how agents and SharePoint agents are paid for without licensing every user.
- Know which Copilot features an admin can turn on or off, such as web search grounding, specific app experiences and plugin or connector availability, and that some controls are tenant-wide while others can be scoped to groups.
- Researcher is the deep-research agent: it runs multi-step reasoning across your work data and the web to produce a sourced, structured report. Reach for it when the task is investigation and synthesis rather than a quick answer.
- Analyst is the data agent: it reasons over raw data such as spreadsheets, writing and running Python to produce analysis and visualisations. It is the answer when the question is about numbers rather than documents.
- Custom agents cover the cases neither built-in agent does - a scoped assistant over a particular set of documents, a departmental helpdesk, a process with defined steps - built in Copilot Studio or as a declarative agent over SharePoint content.
- Assign Copilot licences in the Microsoft 365 admin center, directly or through group-based licensing, and expect a short delay before the experiences appear. Licensing is the first thing to check when a user reports that Copilot is missing.
- Set up and monitor pay-as-you-go billing by linking an Azure subscription and creating a billing policy that scopes which users or sites may consume metered capacity, then watch consumption so costs do not run ahead of expectations.
- Monitor usage and adoption from the Microsoft 365 admin center reports and Copilot Analytics, which show active users, usage by app and adoption trends. Low adoption after a rollout is a change-management finding, not a licensing one.
- Manage prompts as an admin capability: prompts can be saved, shared with the organisation, scheduled to run, and deleted, which is how good prompts spread across an organisation instead of being reinvented by every user.
- Configure user access to agents so only the intended people can use them, through the agent's own sharing settings and the Microsoft 365 admin center controls over which agents are available in the tenant.
- Create an agent from the appropriate surface: a declarative agent over selected SharePoint content for a knowledge assistant, or Copilot Studio when the agent needs topics, actions and connections to other systems.
- Understand the agent approval process: agents submitted for organisation-wide availability go through admin review before they appear for users, which is what prevents an unreviewed agent reaching the whole tenant.
- Monitor agents across their lifecycle using the Microsoft 365 admin center and the Power Platform admin center: usage, operational insights, which agents exist, who owns them, and retiring the ones nobody uses. Ungoverned agent sprawl is the predictable failure mode.
AB-900: M365 Copilot and Agent Administration exam tips
- Data protection and governance is the largest area at 35-40%, and most of it is Microsoft Purview. Learn what each Purview component is for - Information Protection, DLP, Insider Risk Management, Communication Compliance, DSPM for AI, Data Lifecycle Management - because the questions are scenarios matched to a tool.
- The single most important Copilot fact on this exam: Copilot only ever sees what the signed-in user can already see. It does not bypass permissions, so it exposes existing oversharing rather than creating new access. Many answers follow from that one sentence.
- Know which admin center owns which object. Users and licences in the Microsoft 365 admin center, mailboxes and distribution groups in Exchange, sites and libraries in SharePoint, teams and channels in Teams, identity and conditional access in Microsoft Entra, compliance in Purview, and agents in both the Microsoft 365 and Power Platform admin centers.
- Do not confuse sensitivity labels with retention labels. Sensitivity controls who can open content and what they may do with it; retention controls how long it is kept. Both are in Purview and both appear as distractors for the other.
- Learn the DSPM for AI story properly. It is the newest part of the blueprint and the answer whenever a scenario asks how to discover what AI is doing with organisational data or how to find risky prompts.
- For oversharing questions the sequence is: find it with a SharePoint data access governance report, then act on it with SharePoint Advanced Management - restricted access control and restricted content discovery.
- Distinguish the two Copilot billing models, because both appear: a fixed monthly per-user licence for the built-in Copilot experiences, and pay-as-you-go metered consumption billed to an Azure subscription for agents and SharePoint agents.
- This is a fundamentals exam, so questions ask you to identify, understand and compare rather than to configure. Learn what each feature is for and when you would reach for it, in preference to memorising menu paths.
Study guide FAQ
What score do I need to pass AB-900?
A score of 700 or greater is required to pass, on Microsoft's scaled 1-1000 range, and it is not a simple percentage of questions answered correctly. Microsoft does not publish a fixed question count or duration for AB-900 on the study guide, and both can vary between forms. There is no penalty for a wrong answer, so answer everything.
Do I need to be a Microsoft 365 administrator already?
Not necessarily, but you need familiarity rather than only theory. Microsoft expects you to know the core services, security, identity and access, data protection and governance, plus Copilot and agents, and to be familiar with the admin centers for Exchange Online, SharePoint, Teams, Microsoft Entra and Microsoft Purview. Time spent clicking through those consoles in a trial tenant is the most efficient preparation.
Can Microsoft 365 Copilot see data a user is not allowed to see?
No. Copilot runs in the context of the signed-in user and honours their existing Microsoft 365 permissions, so it can only ground answers in content that user could already open. This is why oversharing is such a large part of the exam: Copilot does not create new access, but it makes pre-existing over-permissive sharing far easier to stumble into, which is what data access governance reports and SharePoint Advanced Management exist to fix.
Is my organisation's data used to train the AI models?
No. Prompts, responses and the organisational data accessed through Microsoft Graph are not used to train the foundation models, and the data stays within the Microsoft 365 service boundary under its existing compliance commitments. Expect at least one question that turns on this.
What is the difference between Copilot, Researcher, Analyst and a custom agent?
Microsoft 365 Copilot is the built-in assistant inside the Office apps and Teams, grounded in your Graph data. Researcher is an agent for multi-step investigation across work data and the web, producing a sourced report. Analyst is an agent that reasons over raw data, writing and running Python to analyse and visualise it. Custom agents are ones you build for a specific scope or process, in Copilot Studio or as a declarative agent over SharePoint content.