Study Guide

Licenses map users to services. Microsoft 365 E3 includes core productivity services such as Exchange Online, SharePoint, and Teams, while E5 adds advanced capabilities such as Microsoft Defender for Office 365, eDiscovery, and compliance analytics. Two users with identical group memberships can see different features purely because of their license.

Licenses can be assigned to individual users or to groups. Group-based licensing in the Microsoft 365 admin center works with security groups, mail-enabled groups, and Microsoft 365 groups, and gives every direct member the license. It doesn't support nested groups: only members of the first-level group are licensed. Users without a usage location inherit the tenant's location, and because some services aren't available everywhere, usage location matters. When moving a user between licensed groups, add them to the new group before removing them from the old one, or they briefly lose licensed services.

Assign licenses on Billing > Licenses (choose a product, then up to 20 users or groups at a time) or on Users > Active users. Turn apps and services on or off removes individual services from an assignment. Problems such as no available licenses, conflicting service plans, or an invalid usage location appear on the product's Errors & issues tab, where Reprocess retries the assignment. Assigning licenses requires at least the License Administrator or User Administrator role.

Removing a license has consequences. Exchange Online data is held for 30 days and then deleted, and the mailbox can't be searched with eDiscovery while it's unlicensed. OneDrive files stay until the user account itself is deleted.

Every tenant starts with an onmicrosoft.com domain. Most organizations add a custom domain such as contoso.com for branded email addresses and sign-in names. Go to Settings > Domains > Add domain. If the registrar supports Domain Connect, Microsoft 365 verifies ownership and adds the DNS records automatically. Otherwise, prove ownership with a TXT record (or an MX record, or a text file uploaded to the website), then add the DNS records the wizard lists for each service. A tenant can hold several custom domains, for example for different brands.

Order matters when email moves to Microsoft 365. After the domain is verified, add users and set up their mailboxes before adding the Microsoft 365 MX record, because once the MX record points to Microsoft 365 all mail for the domain goes there. The company website isn't affected. Adding, changing, or removing domains requires the Domain Name Administrator role.

Other organization-wide settings include the organization profile (name, logo, and contact information that appear across services), security defaults (require MFA for all users, block legacy authentication), and external sharing for SharePoint and OneDrive. The admin center also carries Service health, for incidents and advisories, and Message center, for upcoming changes. Settings that belong to one service, such as mail flow rules or meeting policies, live in that service's own admin center.

User mailboxes are created automatically when a user gets an Exchange Online license. Shared mailboxes give a team a common address such as [email protected]; create them under Recipients > Mailboxes. Grant Full Access so people can open and manage the mailbox, and Send As so replies appear to come from it. Send on Behalf, which shows "user on behalf of mailbox", can't be granted in the Exchange admin center and needs the Set-Mailbox cmdlet. Resource mailboxes represent rooms and equipment and can accept or decline booking requests automatically.

Exchange Online has four group types that can receive mail. Distribution lists only distribute messages. Dynamic distribution groups calculate recipients from membership rules. Mail-enabled security groups distribute messages and can also be granted access to resources. Microsoft 365 groups create a collaboration space with a shared inbox, files, and calendar. Create distribution lists under Recipients > Groups > Distribution list.

A distribution list's Settings tab controls delivery management (by default only people inside the organization can send to it), delegates with Send As or Send on Behalf, moderation that holds messages for approval, and whether people can join or leave without owner approval. Every distribution list needs at least one owner.

Choose the site type from the purpose. Communication sites broadcast information, such as news, policies, and training, to a wide audience from a small group of authors. Team sites support group collaboration and are connected to a Microsoft 365 group, so creating one also provisions a shared document library, group mailbox, and calendar.

The SharePoint admin center's Sites > Active sites page lists communication sites, channel sites, and group-connected sites. From it an admin can create sites, change a site's sharing settings, manage storage limits, register or associate hub sites, and add or remove site admins and group owners through Membership. OneDrive sites aren't listed there, but OneDrive settings such as sync, sharing, and storage are managed in the SharePoint admin center.

Inside a site, document libraries store files with version history, metadata columns, content types, and real-time co-authoring. Lists store structured data such as an asset inventory and can trigger Power Automate flows. Folders organize a library's files hierarchically and can be given unique permissions when part of a library needs tighter access.

Every site starts with default SharePoint groups. Visitors get the Read permission level, Members get Edit, and Owners get Full Control. For team sites, Microsoft recommends managing access through the connected Microsoft 365 group; for communication sites, use the built-in SharePoint groups.

Permission levels bundle individual permissions. Read lets people view pages and items and download documents. Contribute adds viewing, adding, updating, and deleting items and documents. Edit adds managing the lists and libraries themselves. Design adds editing pages and applying themes. Full Control contains every permission and, like Limited Access, can't be changed. Limited Access can't be assigned directly: SharePoint adds it to the parent site or library when someone is given access to a single item, so the page can render. View Only and Restricted Read are narrower variants of Read.

Permissions inherit by default: libraries and lists inherit from the site, and folders and files inherit from their library. To give part of a site different access, stop inheriting permissions at that level and assign unique permissions. Sharing a file with someone who doesn't have access also stops inheritance on that file automatically, which is one way unique permissions accumulate across a site.

A team brings people, content, and tools together and is backed by a Microsoft 365 group, so membership changes to the group sync to the team. Teams can be private (invite only) or public (anyone in the organization can join). Channel files are stored in SharePoint.

Channels come in three types. Standard channels are open to every team member. Private channels limit membership to a subset of the team. Shared channels can include people who aren't in the team, including external participants through B2B direct connect, and can be shared with other teams. Private and shared channels each get their own SharePoint site. Guests can take part in standard and private channels but not in shared channels.

Policies control features. Meeting policies cover options such as recording, transcription, and anonymous join; messaging policies cover chat features such as GIFs and message editing; app permission policies decide which apps users can use. Set the Global (Org-wide default) policy first so it covers most users, then create custom policies for exceptions. A user has one effective policy of each type: a direct assignment wins, then the group assignment with the highest ranking, then the Global policy. Policy packages assign a preset set of policies to users with similar roles.

Traditional security trusted anything inside the network perimeter. Modern attacks use identity compromise, phishing, and session hijacking that don't depend on location, so Zero Trust treats every request as untrusted until verified: "never trust, always verify". Verification continues throughout a session rather than happening once at sign-in.

Verify explicitly means every access request is authenticated and authorized using all available signals, such as identity, device health, location, behavior, and risk. In Microsoft 365 that's Conditional Access, MFA, and risk evaluation from Microsoft Entra ID Protection. Use least privilege access means users and workloads get only the access they need, for the shortest time required, through role-based access control and just-in-time elevation with Privileged Identity Management. Assume breach means designing as if attackers are already inside: segment access, monitor continuously, and contain threats fast with tools such as Microsoft Defender XDR and Microsoft Sentinel.

Zero Trust isn't a product or a one-time configuration. Microsoft's training describes an iterative rollout: assess posture with Microsoft Secure Score and Microsoft Purview Compliance Manager, protect identities, enforce endpoint compliance with Intune, classify and protect data with Purview, monitor and respond to threats, and educate users, for example with attack simulation training.

Authentication proves who someone is; authorization, which follows it, decides what that identity can do. In the Microsoft identity platform, OpenID Connect handles authentication and OAuth 2.0 handles authorization, and Microsoft Entra built-in roles, Azure RBAC, and Exchange RBAC are further authorization systems.

Administrative authorization uses role-based access control. Global Administrator controls every setting and should be held by very few people, while service roles such as Exchange Administrator, SharePoint Administrator, Teams Administrator, and User Administrator are scoped to one area. Custom roles fine-tune permissions further. Everyday users are authorized mostly through group membership and resource-specific permissions: joining a Microsoft 365 group gives access to its mailbox, calendar, and site, and SharePoint grants Visitor, Member, or Owner access at the site, library, folder, or file level.

Advanced authorization moves beyond allow-or-deny. Conditional Access applies rules based on sign-in risk, device compliance, and location. Privileged Identity Management grants admin roles just in time for a limited period. Access packages in entitlement management bundle access to several resources behind an approval workflow with an expiration. Sensitivity labels restrict actions such as printing, copying, or forwarding on the content itself, wherever it goes.

Authentication verifies a user's identity before access. Microsoft Entra ID supports methods that differ in where they can be used: as a primary (first-factor) method, as a second factor for MFA, and for self-service password reset (SSPR). A password is primary only. SMS and Authenticator push notifications can be primary, MFA, and SSPR methods. Voice call and software OATH tokens are MFA and SSPR methods but not primary. Passkeys (FIDO2) and Windows Hello for Business are primary and MFA methods but can't be used for SSPR.

Traditional MFA with SMS, email codes, or push approvals is far better than a password alone but can still be phished remotely. Microsoft recommends phishing-resistant methods: Windows Hello for Business, Platform Credential for macOS, synced passkeys, FIDO2 security keys, passkeys in Microsoft Authenticator, and certificate-based authentication. MFA itself requires two or more forms of verification, so a stolen password alone isn't enough.

SSPR lets users reset their own passwords after proving their identity with a method they registered in advance, which cuts helpdesk calls. Microsoft Entra ID Protection adds risk-based checks on top of any method.

Identities are cloud-only (created in Entra ID) or hybrid (synced from on-premises Active Directory with Microsoft Entra Connect Sync or Cloud Sync). Hybrid organizations choose password hash synchronization (simplest, most resilient, commonly recommended), pass-through authentication (an on-premises agent validates passwords in real time and no hash is stored in the cloud), or federation with a provider such as AD FS (most infrastructure, for tight control of sign-in).

Phishing is the most common way in, so protection is layered. Anti-phishing uses spoof intelligence and impersonation protection to catch senders pretending to be trusted domains or executives. Safe Links, part of Microsoft Defender for Office 365, scans URLs during mail flow and rewrites them so each click is checked again at the time of click; in Teams, links are checked at click time without rewriting. Anti-malware policies scan attachments and can block risky file types such as .exe or .js, and anti-spam filtering and mail flow rules handle unwanted and suspicious mail, with quarantine for review.

Some threats are recognized only after delivery. Zero-hour auto purge (ZAP) retroactively detects and neutralizes phishing, spam, and malware already delivered to cloud mailboxes, quarantining malware and high confidence phishing by default, and can also act on malicious Teams chat messages. ZAP works for all cloud mailboxes, not on-premises ones.

Investigation and intelligence tools sit in the Microsoft Defender portal. Threat Explorer (Defender for Office 365 Plan 2) investigates email threats in near real time: pivot by sender, subject, URL, or file hash, see delivery and click status, and remove malicious messages. Plan 1 has Real-time detections, which lacks Explorer's extra capabilities. Threat analytics provides reports from Microsoft security researchers on active threats and threat actors, showing related incidents and exposure in your environment and recommending mitigations. Microsoft's threat intelligence draws on very large volumes of signals across its cloud, endpoint, and identity services, and updates protections as new campaigns appear.

Defender XDR is a pre- and post-breach defense suite. Its main services map to what they protect: Defender for Endpoint (devices: endpoint detection and response, antivirus, attack surface reduction, automated investigation), Defender for Office 365 (email and collaboration: phishing, business email compromise, malicious links and files), Defender for Identity (identities, using on-premises Active Directory signals to detect attacks such as pass-the-hash and lateral movement), and Defender for Cloud Apps (SaaS apps: shadow IT discovery, app governance, data exfiltration control). It also draws on Microsoft Entra ID Protection, Defender Vulnerability Management, and Purview data loss prevention and insider risk signals.

The value is the cross-product layer. Alerts from different products that belong to one attack are joined into a single incident in the Microsoft Defender portal (security.microsoft.com), so analysts see how the attack entered and what it touched. Automated actions heal affected mailboxes, devices, and identities. Advanced hunting lets analysts query 30 days of raw data across Endpoint, Office 365, Identity, and Cloud Apps.

Automatic attack disruption uses high-confidence correlation across workloads to contain an attack in progress, for example by containing a compromised device or disabling a user, which limits lateral movement. Security teams stay in control: they can undo any automatic action and exclude critical assets. Microsoft Purview isn't part of Defender XDR, but the two work together for data security investigations and insider risk.

Microsoft Entra ID is a cloud-based identity and access management service that authenticates users and enforces policy for users, devices, apps, and resources. Every Microsoft 365 tenant is automatically a Microsoft Entra tenant, with an initial onmicrosoft.com domain and any custom domains added. It provides authentication and MFA, single sign-on, group and role management, Conditional Access, and integration with thousands of applications. Admins manage it in the Microsoft Entra admin center or through Microsoft Graph.

Around Entra ID sits the Microsoft Entra product family. Microsoft Entra ID Governance automates access requests, assignments, access reviews, and the identity lifecycle. Microsoft Entra ID Protection detects identity risks such as risky sign-ins and compromised users and drives risk-based Conditional Access. Microsoft Entra Verified ID issues and verifies digital credentials based on decentralized identity standards. Microsoft Entra Private Access reaches private apps without a VPN, and Microsoft Entra Internet Access secures access to internet and SaaS resources. External ID covers partners and customers, Workload ID covers apps and services, and Agent ID gives AI agents governed identities.

Features are licensed in tiers. Microsoft Entra ID Free includes security defaults, which enforce MFA registration and block legacy authentication at no extra cost. Conditional Access requires Microsoft Entra ID P1 (included in Microsoft 365 Business Premium). Risk-based Conditional Access requires Microsoft Entra ID Protection, a P2 feature, and Privileged Identity Management requires Microsoft Entra ID Governance licensing.

A Conditional Access policy is an if-then statement: if a user, group, or agent wants to reach a resource under certain conditions, then they must satisfy a control. Signals include the user or group, IP location (named ranges or countries), device platform and state, the target application, real-time and calculated risk from Microsoft Entra ID Protection, and session information from Microsoft Defender for Cloud Apps. Policies are enforced after first-factor authentication, and they aren't meant as a frontline defense against denial-of-service attacks.

Decisions are either block access, the most restrictive, or grant access with one or more requirements: MFA, an authentication strength, a device marked compliant (compliance is defined in Intune), a Microsoft Entra hybrid joined device, an approved client app or app protection policy, a password change, or terms of use. Session controls go further after access is granted: through Defender for Cloud Apps they can monitor activity or block actions such as downloading, copying, or printing, for example on unmanaged devices.

Common policies require MFA for administrators, block legacy authentication protocols that can't do MFA, block or allow specific locations, block risky sign-ins, require organization-managed devices for sensitive apps, and require trusted locations to register security information. Admins find Conditional Access in the Microsoft Entra admin center under Entra ID > Conditional Access, can start from templates, and can run policies in report-only mode before enforcing them. Conditional Access requires Microsoft Entra ID P1; if the license lapses, policies stay in place but can't be edited.

With SSO, Microsoft Entra ID is the identity provider: an app redirects the user to Entra ID, Entra ID verifies the user's work credentials, and the app trusts that result. Apps no longer keep their own usernames and passwords, and users don't have to sign in again until their session expires or risk changes.

The benefits run both ways. Users manage fewer credentials and sign in less often, so they're less likely to reuse weak passwords or fall for phishing. Admins apply consistent policies such as Conditional Access and MFA at one central point, control access to every app from one place, and see fewer password reset requests.

Entra ID supports several SSO options. Federation-based SSO with SAML 2.0 or OpenID Connect gives the richest integration and suits cloud and modern apps. Password-based SSO stores credentials and replays them for apps that only have a username and password form, often on-premises apps published with Application Proxy. Linked SSO only adds an app link to user portals during migrations, and disabled SSO means users sign in to the app separately. Users find their assigned apps in the My Apps portal. In hybrid environments, users on domain-joined or Entra-joined Windows devices get SSO to Microsoft 365 after signing in to Windows.

A user object in Microsoft Entra ID represents one identity, with a user principal name, credentials, licenses, roles, and group memberships. Users can be employees, external guests, or accounts that services and automation use. Attributes such as department, job title, and location organize users and can drive automation.

Group types have different purposes. Security groups grant access to resources such as SharePoint sites and can contain devices, which Intune uses. Microsoft 365 groups are for collaboration and provision a shared mailbox, calendar, SharePoint library, and Planner, and Teams uses them for membership. Mail-enabled security groups grant access and receive email, but can't use dynamic membership or contain devices. Distribution groups only send email, and dynamic distribution groups calculate recipients from filters defined in the Exchange admin center. Microsoft 365 groups can't be members of security groups or distribution groups.

Dynamic membership in Microsoft Entra ID adds and removes members automatically from rules on user or device attributes, such as Department equals Sales. It's available for security groups and Microsoft 365 groups, members can't be added manually, and each unique user in dynamic groups needs a Microsoft Entra ID P1 license. As a best practice, assign admin roles to groups rather than individuals, so joining or leaving a team grants or removes the role automatically, and review memberships regularly.

The Microsoft Entra sign-in logs (Entra ID > Monitoring & health > Sign-in logs, at least Reports Reader) record every sign-in attempt: who signed in, from which app, device, location, and IP address, whether MFA was satisfied, which Conditional Access policies applied, and the result with a failure reason. Entries can't be changed or deleted. Filter by user, app, IP address, or date to find the event.

For MFA failures, confirm the failing policy in the sign-in logs and check whether the user's registered methods meet it. A phishing-resistant policy blocks users who only have weaker methods, and a lost or reset phone means re-registering. Authenticator time-based codes fail when the device clock is wrong. For Conditional Access denials, use the What If tool (Entra ID > Conditional Access > Policies > What If): enter the identity, target resource, device platform, and client app, and it lists the policies that would apply, with their controls, and those that wouldn't, with the first unmet condition. Then check the device's compliance in Intune and any alerts in Defender for Endpoint.

Risky sign-ins come from Microsoft Entra ID Protection, which detects signals such as atypical travel, anonymous IP addresses, unfamiliar sign-in properties, password spray, and leaked credentials. Full detection details need Microsoft Entra ID P2. Filter sign-ins by risk level, and respond by requiring MFA, forcing a password reset, or blocking access. Other common causes are location-based policies (check the Location and IP fields) and legacy clients using POP, IMAP, or SMTP AUTH (check the Client app field), which fail where legacy authentication is blocked.

Every 24 hours, Microsoft compares your identity configuration with recommended practices and recalculates the score. Find it in the Microsoft Entra admin center under Entra ID > Identity Secure Score, or under Recommendations with the Security filter. It's available to free and paid tenants, although some recommendations need a paid license. Recommendations include requiring MFA for administrative roles, ensuring all users can complete MFA, blocking legacy authentication, enabling self-service password reset, protecting users with sign-in and user risk policies, and using least privileged administrative roles.

Some recommendations are scored all-or-nothing and others as a percentage. For example, if protecting all users with MFA is worth up to 10.71% and 5 of 100 users are protected, you get about 0.53%. Identity Secure Score is the identity category of Microsoft Secure Score, which also has data, devices, infrastructure, and apps categories, and the recommendations are the same.

Each improvement action has a status. To address means it's needed (including partly done actions). Planned means concrete plans exist. Risk accepted means you've chosen not to act: no points are awarded and the action leaves the list. Resolved through third party or Resolved through alternate mitigation means another tool covers it, and the points are awarded. The score measures how much of the recommended protection you've adopted, not the probability of a breach, so prioritize high-importance recommendations over a target number. Viewing improvement actions needs at least Service Support Administrator; updating their status needs a more privileged role.

Audit log search in the Microsoft Purview portal (the Audit solution) covers activity across Exchange, SharePoint, OneDrive, Teams, Microsoft Entra ID, and more. It's on by default for enterprise organizations. Searching needs the Audit Logs or View-Only Audit Logs role. Filter by date range (up to 180 days per search), activity (friendly names or exact operation names such as FileDownloaded), record type, users, file, folder, or site, and workload. Searches run as jobs that continue after the browser closes, and results export to CSV. In PowerShell, Search-UnifiedAuditLog in Exchange Online PowerShell does the same, and the Office 365 Management Activity API downloads audit data programmatically.

Retention depends on the user's license. Records for users with non-E5 licenses (Audit Standard) are kept 180 days; the default rose from 90 days for logs created on or after October 17, 2023. For users with Office 365 E5, Microsoft 365 E5, or an equivalent add-on, Microsoft Entra ID, Exchange, and SharePoint records are kept one year by default, and audit log retention policies can keep other services' records up to a year. Core service records typically appear 60 to 90 minutes after the event.

Microsoft Entra has its own logs in the Microsoft Entra admin center. Audit logs record changes to the tenant, such as user and group management, and sign-in logs record sign-in attempts. Typical investigations: filter FileDownloaded or FileAccessed on a sensitive site to trace data exfiltration, track assignments of privileged roles such as Global Administrator, and trace changes to Teams policies or mailbox permissions. Exporting to Microsoft Sentinel or another SIEM enables alerting and longer-term analysis.

Fewer people with standing privileged access means less chance of an attacker or a mistake misusing it. PIM, a Microsoft Entra ID service, manages access to Microsoft Entra roles, Azure resource roles, and groups, and resources in Microsoft 365 and Intune. It requires Microsoft Entra ID Governance licensing.

Assignments are eligible or active. An eligible user must activate the role before using it, by completing MFA, giving a business justification, or getting approval from designated approvers, and the activation lasts only up to a configured maximum before the role turns off. An active assignment works with no action. Once activated, eligible and active assignments give the same access; eligibility simply means the access isn't there all the time. Either type can be permanent or time-bound with start and end dates, and users can request extensions or renewals, which a Global Administrator or Privileged Role Administrator approves.

PIM also sends notifications when privileged roles are activated, runs access reviews so users keep only roles they still need, keeps downloadable audit history for internal and external audits, and prevents removing the last active Global Administrator or Privileged Role Administrator assignment. Only a Privileged Role Administrator or Global Administrator can manage PIM assignments for other admins on Microsoft Entra roles.

To use Microsoft Entra ID for sign-in and access, an application must be registered. Registration creates the application object in the app's home tenant: its identity configuration, including supported account types (single tenant or multitenant), redirect URIs, the permissions it needs, and its credentials (client secrets or certificates, which admins must protect and rotate). Manage these on the App registrations page. Registering an app in the admin center also creates its service principal in the home tenant.

A service principal is the app's local representation in a tenant. It defines what the app can do there, who can access it, and which resources it can reach. The application object is the template, and a service principal is created in every tenant where the app is used, for example when an administrator of another organization consents to a multitenant app. Manage service principals on the Enterprise applications page, where you can review permissions, user consent, and sign-ins, assign users, and configure SSO for SaaS apps. Managed identities are also service principals, without an application object.

Consent decides what apps can access. By default, users can consent to permissions that don't require admin consent, such as an app reading their own mailbox. To reduce the risk of malicious apps, Microsoft recommends allowing user consent only for apps from verified publishers (Enterprise apps > Consent and permissions > User consent settings). Apps that require user assignment always need admin consent.

Microsoft Purview is Microsoft's platform for securing, governing, and managing compliance for data wherever it lives, managed in the Microsoft Purview portal. Its solutions share capabilities such as classifiers, sensitivity labels, and data and activity explorers.

Information Protection classifies and protects content with sensitivity labels that can encrypt and mark files, emails, meetings, and containers. Data Loss Prevention (DLP) stops sensitive data from being shared inappropriately: policies inspect content across Exchange, SharePoint, OneDrive, Teams, Office apps, Windows and macOS devices (endpoint DLP), non-Microsoft cloud apps, on-premises file shares, Fabric and Power BI, and Copilot, and can warn with a policy tip, block with an override and justification, or block outright. Monitored activity goes to the audit log and Activity explorer.

Insider Risk Management correlates signals to detect malicious or inadvertent risks from people who already have access, such as IP theft and data leaks. Policy templates include Data theft by departing users, Data leaks, Security policy violations, Risky AI usage, and Risky Agents. Users are pseudonymized by default, and cases can escalate to eDiscovery (Premium). Communication Compliance checks messages in email, Teams, Viva Engage, Copilot interactions, and third-party channels for regulatory (such as FINRA) and conduct violations such as harassment, threats, or leaks of confidential information, with reviewers acting on matches.

Data Security Posture Management (DSPM) for AI is the central place to see and secure AI use across Copilot, agents, enterprise AI apps, and third-party AI sites: activity insights, one-click policies, recommendations, and data risk assessments for oversharing (a default assessment runs weekly for the top 100 SharePoint sites by usage). Learn now labels it classic and points to a new Data Security Posture Management experience. Data Lifecycle Management (formerly Microsoft Information Governance) keeps what you need and deletes what you don't through retention policies and labels, plus archive mailboxes, inactive mailboxes, and PST import.

A sensitivity label is a customizable, clear-text, persistent tag stored in a file's or email's metadata, so it stays with the content wherever it's saved or sent. Each item can have one sensitivity label, alongside an optional retention label. A label can simply classify content, or it can enforce protection: encryption that controls who can open content and what they can do with it, and content markings such as headers, footers, and watermarks (watermarks don't apply to email).

Use cases go beyond files. Labels with the Meetings scope protect calendar items, Teams meetings, and chat. Labels with the Groups & sites scope, available after enabling sensitivity labels for containers, protect Teams, Microsoft 365 groups, SharePoint sites, Viva Engage communities, and Loop workspaces by controlling privacy, external user access and sharing, and access from unmanaged devices, without labeling the items inside. Labels can set the default sharing link type for SharePoint and OneDrive, extend SharePoint permissions to downloaded files, filter eDiscovery searches, and protect content in third-party apps through Microsoft Defender for Cloud Apps. Copilot and agents recognize labels and honor their protection.

Labels reach users through label policies, which decide which labels users see and options such as mandatory labeling, a default label, and requiring justification to lower a label. Label order sets priority: least restrictive at the top, most restrictive at the bottom. Labels can be applied manually, recommended to the user with a policy tip, or applied automatically when content matches conditions such as sensitive information types.

Classification tells you what sensitive data you have and where, so protection can target it. Purview offers three approaches: manual classification by users and admins; automated pattern matching using keywords or metadata, sensitive information types, document fingerprinting, and exact data match; and trainable classifiers. Data classification scans sensitive and labeled content before any policy exists, so you can see the landscape before designing policies.

Sensitive information types (SITs) find data such as credit card, passport, or Social Security numbers using patterns like regular expressions, nearby keyword evidence, validation checks, and confidence levels. Microsoft provides many built-in SITs, and you can create custom ones. Because a pattern matches anything shaped the same way, generic SITs can produce false positives. Exact data match (EDM) SITs instead compare content to a refreshable database of your actual values, such as real employee IDs, which cuts false positives; the database can hold up to 100 million rows and isn't shared with anyone, including Microsoft.

Trainable classifiers recognize content that has no reliable pattern, such as contracts, source code, resumes, or merger documents, by learning from examples. Pretrained classifiers are ready to use. Custom classifiers are trained with examples that are and aren't in the category, then tested and published; they support English only, can't be retrained after publishing, and don't work on encrypted items. Classifiers can be used for auto-labeling, DLP, retention label auto-apply policies, and Communication Compliance (Microsoft-provided classifiers only).

Retention settings keep content for a period, delete it after a period, or retain and then delete it. Use a retention policy for the same settings across a location, such as all mailboxes or specific SharePoint sites. Use a retention label when individual items need different settings; labels can be published for users to apply or applied automatically using keywords, sensitive information types, or trainable classifiers. Label settings travel with an item moved within the tenant, and only labels support retention periods that start when an item is labeled or when an event occurs, disposition review before deletion, and marking items as records.

Content is retained in place. If a user edits or deletes an item under retention, a copy is kept: SharePoint and OneDrive copies go to the site's Preservation Hold library (which counts toward storage), Exchange copies go to the Recoverable Items folder, and Teams, Viva Engage, Copilot, and AI app messages go to a hidden SubstrateHolds folder in Recoverable Items. Users generally don't notice.

An item can be under several retention policies and one retention label. The principles of retention resolve conflicts in order: retention wins over deletion; the longest retention period wins; for deletions, explicit wins over implicit, so a retention label's delete action beats a retention policy's; and finally the shortest deletion period wins. For example, a policy that deletes after three years and a label that retains for five years keep the item for five years before deletion.

Copilot coordinates three components: large language models, content in Microsoft Graph that the user can access, and the Microsoft 365 apps. When a user enters a prompt, Copilot grounds it with relevant Graph data, sends the grounded prompt to the LLM, and returns the response to the app. Data is encrypted in transit, customer data stays within the Microsoft 365 service boundary, and existing security, compliance, and privacy policies still apply. Learn now calls the product Microsoft Copilot, while the exam outline uses Microsoft 365 Copilot; there are no changes to its security or privacy.

Access is permission-trimmed. Copilot only surfaces data the user has at least view permission to, so if a user can't open a file, Copilot can't use it for them. Being inside the service boundary doesn't give Copilot tenant-wide visibility. This makes existing permissions the key control: overly broad sharing means Copilot can surface content to more people.

Prompts, responses, and Graph data aren't used to train foundation LLMs. Interactions are stored as the user's Copilot activity history, encrypted and under the same commitments as other Microsoft 365 content; admins can find them with Content search and govern them with Purview retention, and users can delete their own history. Copilot can reach beyond Microsoft 365 through Graph connectors, whose data returns only to users with permission, and through agents, which admins allow or block in Integrated apps in the Microsoft 365 admin center. Web grounding sends a generated search query to Bing rather than the full prompt.

Microsoft Graph is the data layer across Microsoft 365: emails, chats, meetings, calendar, contacts, and documents. Copilot combines that content with the user's working context, such as the meeting they're in, recent email exchanges on a topic, and last week's chats, to produce relevant answers. Graph ranks results using signals such as recent file activity, sharing, and relationships between people and content, and access to Graph data is gated by role-based access control.

The semantic index, generated from Graph content for users with a paid Copilot license, represents content as vectors so that similar meanings sit close together. That lets Copilot recognize word forms and synonyms and find related material even when the exact words differ, alongside traditional keyword search. It honors each user's access boundary. When a user attaches a SharePoint library or folder to a prompt, Copilot scopes grounding to it and uses the library's column metadata to rank results.

Because Graph is what Copilot draws on, content governance shapes output. Overly permissive sharing, unstructured libraries, and stale files lead to irrelevant or excessive results, while clear structure, meaningful metadata, and deliberate sharing make responses faster, more relevant, and safer.

In Microsoft 365, permissions are the first control: Copilot uses the same access model as every other service. SharePoint tools such as Restricted Content Discovery and SharePoint Advanced Management help limit what content Copilot can reach while oversharing is fixed.

In Microsoft Purview, sensitivity labels show users the sensitivity of content, and when a label applies encryption, a user needs both VIEW and EXTRACT usage rights for Copilot to return the data. Copilot doesn't return S/MIME-protected email and can't use password-protected documents unless they're already open. DLP policies for the Microsoft 365 Copilot and Copilot Chat location can prevent Copilot from using files and emails with chosen labels (they may still appear as citations, but their content isn't used), stop Copilot from responding to prompts that contain chosen sensitive information types (preview), block external web search for such prompts, or exclude email from external senders (preview). That location is only in the Custom policy template and doesn't support admin units, and DLP checks typed prompt text, not uploaded files. Prompts and responses are captured in the audit log, can be retained or deleted with retention policies, found with eDiscovery, and checked with Communication Compliance.

Identity and threat controls also apply. Copilot honors Conditional Access policies and MFA, so a device or location restriction on SharePoint also limits Copilot. Insider Risk Management's Risky AI usage template detects risky prompts such as prompt injection attempts, and those insights are integrated into Microsoft Defender XDR.

Microsoft's Responsible AI Standard is built on six principles. Fairness and inclusiveness mean AI systems treat everyone fairly and don't affect similar groups of people differently; for example, they make the same recommendations for people with similar circumstances or qualifications. Reliability and safety means systems operate consistently, respond safely to unexpected conditions, and resist harmful manipulation. Privacy and security means protecting personal and business data and complying with privacy laws. Transparency means people can understand how an AI system reaches its outputs. Accountability means the people who design and deploy AI remain responsible for it, AI isn't the final authority on decisions that affect people's lives, and humans keep meaningful control.

In Copilot these principles become product behavior. Harm filters address hate and fairness, sexual, violence, and self-harm content in prompts and responses. Protected material detection flags copyrighted text and code with licensing restrictions. Jailbreak and cross-prompt injection classifiers block high-risk prompts before the model runs. Copilot also restricts AI from making inferences about an employee's performance, attitude, emotional state, or personal characteristics from workplace communications. Answers cite their sources so users can verify them.

Generative AI responses aren't guaranteed to be 100% factual, so Copilot is designed to produce drafts and summaries that people review before sending or acting on them. Microsoft doesn't claim ownership of Copilot output, and its Copilot Copyright Commitment defends commercial customers sued over output, provided they used the built-in guardrails and content filters.

Compliance Manager helps you inventory data protection risks, implement controls, keep up with regulations, and report to auditors. An assessment groups the controls for one regulation, standard, or policy; Compliance Manager offers over 360 regulatory templates and custom templates. Controls are Microsoft managed (implemented by Microsoft for its cloud services), yours (customer managed), or shared.

Improvement actions are the recommended steps to meet controls, such as enabling auditing or deploying DLP for personal data. You can assign them to people, store evidence and notes, and track status; status changes appear on the dashboard within 24 hours. Every tenant starts with an initial score from the Microsoft 365 Data Protection Baseline assessment, which draws on NIST CSF, ISO, FedRAMP, and GDPR.

The compliance score awards points for completed actions, weighted by risk. Actions are mandatory (can't be bypassed, like an enforced password policy) or discretionary (rely on users), and preventative, detective, or corrective. A preventative mandatory action is worth 27 points, preventative discretionary 9, detective or corrective mandatory 3, and detective or corrective discretionary 1. Recommendations aren't a guarantee of compliance, so you still validate your own controls.

Data explorer (Microsoft Purview portal > Information Protection > Explorers > Data explorer) lists items that have a sensitivity label or retention label, or that match a sensitive information type or trainable classifier, across locations such as Exchange, SharePoint, and OneDrive. Filter by label, classifier, or information type, then drill into locations and items. It's especially useful before deploying new policies, to see what's really out there.

Access is tightly controlled because data explorer can reveal file contents regardless of item permissions. Roles such as Compliance Administrator, Security Administrator, and Compliance Data Administrator open the tab, but seeing items needs the Data Explorer List viewer role group, and reading contents (and item names, which can be sensitive) needs Data Explorer Content viewer. The two are independent, and data explorer supports administrative units.

Encrypted documents can't be previewed or downloaded. For sensitivity labels on SharePoint and OneDrive files to appear, sensitivity labels must be enabled for those services. Data explorer answers "what sensitive content exists and where"; Activity explorer answers "what did people do with it".

Policies built from templates (such as Data theft by departing users or Data leaks) score risk indicators and generate alerts with a severity of High, Medium, or Low. Analysts start on the alert dashboard: filter to Needs review, work from the highest severity, and use spotlighted alerts or the Alert Triage Agent to prioritize. Severity can change over time if alerts aren't triaged, and users are pseudonymized by default.

Each alert's details page supports investigation. The Activity explorer tab shows a timeline of the risky activities behind the alert, the Data risk graph connects users and files, and Summarize with Copilot pulls together the triggering event, activity, user attributes, last working date, and top risk factors. Content preview (preview) shows relevant files so analysts can confirm whether sensitive data is involved.

After review, the analyst confirms the alert and creates a new case or adds it to an existing case, or dismisses it, including dismissing false positives in bulk. Cases hold the investigation and can escalate to eDiscovery (Premium). The training's examples of what policies detect include mass downloads of confidential files, sending documents to personal email, and unusual access times, especially combined with HR signals such as a resignation.

Alerts are configured in each DLP policy. When a rule matches, an alert with details such as the user, time, location, data, and rule appears in Microsoft Purview portal > Data loss prevention > Alerts, for Exchange, SharePoint, OneDrive, Teams, devices, on-premises repositories, and Fabric and Power BI. Filter and customize columns, open an alert for its details, events, and metadata, and optionally use Summarize with Copilot or the Alert Triage Agent (preview).

Viewing the dashboard needs the Manage alerts role plus DLP Compliance Management or View-Only DLP Compliance Management, which role groups such as Compliance Administrator, Security Administrator, Security Reader, and the Information Protection groups include. Content preview and matched sensitive content need the Content Explorer Content Viewer role group. How long alerts stay visible depends on the organization's audit log retention.

Responses depend on severity: investigate who shared what and why, notify or educate the user, tune a policy that produces false positives, or escalate. The training describes routing DLP alerts to Microsoft Defender XDR, Microsoft Sentinel, or another SIEM, and triggering automated workflows with Logic Apps or Power Automate.

Communication Compliance policies check email, Teams, Viva Engage, Copilot interactions, and third-party communications with built-in classifiers (such as threats, harassment, and profanity), keywords, and sensitive information types. OCR detects text in images, and linked attachments in Teams can be analyzed. Matches appear on each policy's Pending tab, with usernames pseudonymized by default.

Reviewers examine the message basics and details: keyword highlighting shows the matched terms, and they can see attachments, conversation context, user history of past remediation, pattern-detected notifications for recurring behavior, and translation. Copilot in Microsoft Purview can summarize long messages and meeting transcripts.

Remediation actions are Resolve (optionally marking the match as misclassified), Tag as Compliant, Noncompliant, or Questionable (or custom tags), Notify the sender with a notice template, Escalate to another reviewer, Escalate for investigation (creates an eDiscovery (Premium) case, up to 100 messages at once), run a Power Automate flow such as notifying the user's manager, and Remove message in Teams, which replaces the message with a policy tip. Resolved items move to the Resolved tab.

Where data explorer shows what sensitive content exists, Activity explorer shows what happened to it: labels applied, changed, or removed; files shared, downloaded, or copied to USB; endpoint and network DLP activities; and DLP policy matches. It isn't limited to policy violations, so it also helps find risky patterns before they become incidents.

About 50 filters narrow the view, including date range, activity type, location, sensitivity label, user, client IP, device name, and whether the item is protected. Predefined filter sets cover common investigations, such as Endpoint DLP activities, Sensitivity labels applied, changed, or removed, Egress activities, and DLP policies that detected activities, and you can save your own. Security Copilot (preview) can generate filter sets from natural language.

Activity explorer isn't real time. Its data comes from the unified audit log, so core service activities typically appear 60 to 90 minutes after they happen, and offline devices report activities when they reconnect. To confirm a specific activity independently, search for it in Audit. Access needs explicit membership in an appropriate Purview role or role group.

Start in the Microsoft Purview portal under DSPM for AI, with an account such as Compliance Administrator. Copilot and agent insights need Microsoft Purview Audit, which is on by default for new tenants. Insights about third-party AI sites, such as ChatGPT or Gemini, also need the Microsoft Purview browser extension and devices onboarded to Purview. One-click policies (for example, to discover users sending sensitive information to AI sites) take at least 24 hours to collect data.

Reports and Activity explorer then show AI activity grouped into Copilot experiences and agents, Enterprise AI apps (such as ChatGPT Enterprise), and Other AI apps. Activity explorer events include AI interaction, Sensitive info types, and AI website visit, with the user, app, sensitive information types, and files referenced; prompts and responses are included for people with the right permissions. The Apps and agents page lists AI apps and agents in use, what sensitive data they accessed, and which Purview policies protect them.

To manage what you find, work through recommendations: run data risk assessments for oversharing, create default sensitivity labels, protect labeled items from Copilot and agent processing with DLP, detect risky interactions in AI apps (which feeds Insider Risk Management's risky AI usage signals), and use Compliance Manager templates for AI regulations. Learn now labels this experience DSPM for AI (classic) and points to a new Data Security Posture Management experience; the exam outline uses the DSPM for AI name.

Microsoft retired the classic Content Search, eDiscovery (Standard), and eDiscovery (Premium) experiences on August 31, 2025. In the Microsoft Purview portal, all Content Search functionality is part of eDiscovery: a system-generated Content Search case is available to eDiscovery Manager and eDiscovery Administrator role group members, and any case can hold searches. Cases are the central organizing unit.

eDiscovery searches Exchange Online, Teams, Microsoft 365 Groups, OneDrive, SharePoint, and Viva Engage, with mailboxes and sites in the same search. Use keyword queries (KQL) and conditions, review statistics and samples, place content locations on hold, export results, and search for and delete data such as a phishing email. Premium features with E5 licensing add review sets, advanced indexing, OCR, conversation threading, decryption, and importing external data.

Copilot and other AI app prompts and responses are stored in the user's mailbox. To find them, search that mailbox with the condition Type > Contains any of > Copilot activity, then export the results or add them to a review set. Retention policies and Audit also cover these interactions.

Oversharing happens when content is shared more broadly than intended: Anyone links or sharing with Everyone except external users, Edit access where View would do, or guests given whole-site access for one file. Copilot doesn't create the problem, but because it can surface anything a user can technically reach, it makes oversharing more visible, which is why Microsoft's secure and govern blueprint for Copilot starts with remediating oversharing, then setting up guardrails and meeting regulations.

Discovery tools include SharePoint data access governance reports (site permissions, sharing links, EEEU sharing, sensitivity labels on files), DSPM for AI data risk assessments (a weekly default assessment of the top 100 sites by usage, plus custom assessments with remediation options), sharing reports, and SharePoint Online PowerShell or Microsoft Graph for audits at scale.

While owners fix permissions, Restricted Content Discovery can temporarily keep up to 20,000 SharePoint sites (not OneDrive) out of organization-wide search and Copilot responses, and removes Copilot entry points on them, without changing permissions; use it selectively. Restricted SharePoint Search, an older 100-site allow list that was never a security boundary, is retiring: new enablement is blocked from July 31, 2026. Longer-term prevention comes from sensitivity labels with encryption and default sharing link types, organization-wide sharing limits such as disabling Anyone links, user education and alerts, and SharePoint Advanced Management controls.

Open the SharePoint admin center with SharePoint administrator credentials, expand Reports, and select Data access governance. Reports require SharePoint Advanced Management prerequisites. Organizations with only Microsoft 365 E5 get limited reporting: activity reports (up to 10,000 sites) without snapshot reports or remedial actions, and must enable data collection first.

Snapshot reports show the state at the time they're generated. The site permissions for your organization report shows the broadest exposure across all SharePoint and OneDrive sites, such as sites with thousands of users, guests, or EEEU permissions. The site permissions for users report lists every site a specified user can access and how. The sensitivity labels for files report finds sites holding files with chosen labels, and the special SharePoint groups report pinpoints items made effectively public through EEEU or Everyone. Activity reports track the last 28 days: the sharing links report finds sites with the most new Anyone, organization, or specific-people links, and the EEEU report finds sites where content was shared with all internal users. Microsoft suggests snapshot reports quarterly and activity reports monthly.

Because IT often can't inspect file-level details for compliance reasons, site access reviews delegate remediation. From a supported report (sharing links, EEEU, or the permissions baseline), select up to 100 sites in the web view (more with PowerShell) and each site owner receives an email tailored to the issue found. Site access reviews support SharePoint sites only, not OneDrive.

SharePoint Advanced Management (SAM) is managed mainly in the SharePoint admin center. An organization with a qualifying base subscription gets SAM capabilities that support Copilot as soon as at least one user is assigned a Microsoft Copilot license; otherwise it can buy the SAM Plan 1 add-on, which some features, such as restricted site creation, require. Learn's training describes SAM as a paid add-on, but current product docs say it's included with Copilot licenses.

For content sprawl, SAM offers the inactive SharePoint sites policy (detect inactive sites and notify owners), the site ownership policy (minimum owners or admins, with notifications), and site attestations (owners confirm a site is still needed and its access is right). For the content lifecycle, it offers catalog management, change history reports covering the last 180 days, recent site property changes, and restricting which apps can create sites. For oversharing, it offers Conditional Access through authentication contexts, block download policies, Restricted Content Discovery, DAG reports with site access reviews, site policy comparison, content management assessment, app and agent insights, and OneDrive access restrictions.

Restricted access control (also called restricted site access or site access restriction) limits a SharePoint or OneDrive site to members of up to 10 Microsoft 365 groups or Microsoft Entra security groups, including dynamic groups. Users outside those groups can't open the site or its content even if they have permissions or a sharing link, and organization-wide search and Copilot honor the restriction. Group membership alone doesn't grant access: a user needs both permissions and membership. An admin first allows site-level access restriction for the organization (Policies > Access control), then adds groups on the site's Settings tab in Active sites or with Set-SPOSite. A group-connected site's own Microsoft 365 group is the default control group, and private and shared channel sites need their own policies.

Microsoft 365 Copilot is built into apps such as Word, Excel, PowerPoint, Outlook, and Teams and into Copilot Chat. It responds to prompts, grounded in data the user can access. Agents extend it for specific domains: they combine instructions, knowledge sources, and actions, so they can answer from particular content, take actions such as updating records, and automate multistep processes. Use Copilot for one-off, user-driven tasks, prebuilt agents for common standardized needs, and custom agents for repeatable, department-specific, or cross-system processes.

There are two build approaches. Declarative agents use Copilot's own orchestrator and models, so they inherit Microsoft 365 security, compliance, and responsible AI standards and need no extra hosting; they're built with Agent Builder, Copilot Studio, or the Agents Toolkit. Custom engine agents bring their own orchestration and models, usually need extra hosting such as Azure, can act proactively, and their owners must handle compliance themselves.

What a user can do depends on the license. Copilot Chat (Basic), included with eligible Microsoft 365 plans, grounds responses in the web, can use work content only when the user uploads it or has it open in Outlook or Teams, includes web-data agents, and can use work-data agents on pay-as-you-go. Microsoft 365 Copilot (Basic) adds standard in-app Copilot without the add-on. Microsoft 365 Copilot (Premium), the add-on license, adds priority access, automatic grounding in Microsoft Graph and Work IQ, and agents that use work data. Learn now calls the product Microsoft Copilot.

The monthly model is a per-user add-on license at a fixed cost, bought on top of a qualifying plan such as Microsoft 365 E3 or E5, a Business plan, or Office 365 E3. Costs are easy to budget and assignment uses familiar tools, including group-based licensing, but money is wasted on licensed users who rarely use Copilot, so regular license audits matter.

Pay-as-you-go lets organizations use Copilot capabilities without a full license, paying only for usage. It applies to Microsoft Copilot Chat (metered agents), SharePoint agents, and the Copilot Retrieval API (preview), and it's off by default. Usage is billed through the Azure subscription linked to a billing policy, so costs appear alongside other Azure consumption. Copilot Studio has its own pay-as-you-go billing, and Copilot Cowork and the Work IQ API use Copilot Credits usage-based billing.

Choose by usage pattern. Pay-as-you-go suits pilots, seasonal or occasional users, and establishing usage patterns before committing to licenses, but spend can spike, so monitor it. Monthly licenses suit steady daily use. For SharePoint, users without Copilot licenses can use SharePoint agents when their billing policy is connected to the SharePoint agents service, so heavy querying of large libraries can raise costs; either way, Copilot only surfaces SharePoint content the user can access.

Copilot settings are in the Microsoft 365 admin center under Copilot > Settings, split into User access, Data access, Copilot actions, and Other settings. The AI Administrator role can make changes without Global Administrator rights; Global Reader can view. Some items are configured directly and others open another admin center. These settings are part of Copilot controls, Microsoft's framework of security and governance, management controls, and measurement and reporting for Copilot and agents.

User access settings include Copilot Frontier (early access to experimental features, no access by default), Copilot pay-as-you-go billing, Microsoft Copilot in admin centers, Microsoft Copilot self-service purchases (Allow, Allow trials only, or Do not allow), and pinning Copilot Chat in Teams, Outlook, and the Copilot app. Data access settings include Agents (who can access agents and which types they can install), AI providers for other large language models, and web search, which is controlled by the Allow web search in Copilot cloud policy in the Microsoft 365 Apps admin center. Copilot actions include Copilot image generation, screen and camera sharing (vision), and a shortcut to Copilot in Teams meetings in the Teams admin center.

Beyond these settings, rollout is shaped by licenses and groups: assign Copilot licenses to pilot groups, and use app-specific controls where they exist. For example, Teams meeting recaps depend on whether transcription is allowed, so restricting transcription restricts recaps. Admins can also remove user access to Copilot Chat entirely.

Researcher is a first-party agent in Microsoft Copilot for tasks that need deeper reasoning: gathering and analyzing information from the web and from work content the user can access, such as files, emails, meetings, and chats, then producing a report with organized sections, visuals, charts, and cited sources. It takes longer by design. For quick summaries, short replies, or light back-and-forth, standard Copilot chat is the better choice.

Typical uses include market or competitive analysis, executive briefings, knowledge discovery across scattered documents and conversations (for example, summarizing root causes of past incidents), condensing long documents while linking back to sources, and pulling a consolidated project view from several apps. Users find it in the Microsoft Copilot app under Agents in Chat.

For the best results, write a specific prompt, decide whether to search workplace resources, the web, or both, and answer Researcher's clarifying questions, which steer the research. Researcher respects existing permissions, policies, and compliance. Researcher and Analyst are part of the core Copilot Chat experience: they follow agent governance generally but aren't controlled by agent-related admin settings.

Analyst works like a virtual data analyst. From natural language prompts it summarizes data, compares segments, identifies trends and anomalies, builds forecasts and what-if scenarios, and creates visuals from spreadsheets and other structured data the user provides or references. In supported experiences it runs Python for complex calculations and shows its analysis steps so users can review the reasoning. Users can keep refining with follow-up requests.

Good fits include business reviews, tracking performance metrics, spotting outliers, demand forecasts, and revenue projections, especially for people who aren't data experts. Analyst respects file permissions, so it only uses data the user can see.

Know its limits. It works best with clean, tabular data; freeform text, screenshots, or handwritten notes may need organizing first. It doesn't replace subject matter experts or dedicated BI tools for large-scale reporting, may miss subtle data inconsistencies, and isn't a real-time processor, so rerun requests when source data changes. For critical decisions, treat results as a draft for human review. Compared with Researcher: Analyst is for structured data and numbers; Researcher is for unstructured information and web research.

Agent Builder in Microsoft 365 Copilot lets information workers create lightweight agents in natural language for themselves or small teams, grounded in organizational content: project FAQ agents, product documentation assistants, and onboarding agents. These agents respect existing permissions and are managed mainly in the Microsoft 365 admin center. If needs grow, an agent can be copied to Copilot Studio.

SharePoint agents help people find and use content in SharePoint sites, libraries, and files. A new site can get a ready-made agent scoped to that site, and users with Edit permission can create custom SharePoint agents that include other sites, libraries, or files as sources.

Copilot Studio is for makers and developers building agents for a department, the whole organization, or external customers, or agents that need multistep logic, approvals, connectors to business systems, autonomous capabilities, and lifecycle management across environments. Examples include customer support agents that create tickets and escalate to a person, IT help desk triage, and sales assistants that work with CRM data. These agents are governed mainly in the Power Platform admin center. Before building a custom agent, check whether Copilot or a prebuilt agent already covers the need, and plan for long-term maintenance.

Microsoft 365 Copilot is an add-on to a qualifying plan, such as Microsoft 365 E3, E5, or E7, Microsoft 365 Business Basic, Standard, or Premium, Microsoft 365 F1 or F3, Office 365 E1, E3, E5, or F3, and several Teams and Exchange plans. The Copilot setup guide in the admin center steps through assigning licenses, and the Copilot License Details diagnostic checks whether a specific user meets the requirements. The Recommendations for Microsoft Copilot licensing setting suggests candidates based on app activity.

Assign licenses to individuals under Users > Active users > Licenses and apps, or on Billing > Licenses. For departments, use group-based licensing so members get the license automatically and lose it when they leave the group, which keeps costs aligned with real users. Make license assignment part of onboarding, and audit assignments regularly to reclaim unused licenses.

Use least-privilege roles: License Administrator to assign licenses, rather than Global Administrator. When Copilot doesn't appear for a licensed user, confirm the license is assigned and has propagated, check that Microsoft 365 Apps are on a current build, check Service health for Copilot incidents, and consider network access to required endpoints.

Setup happens in the Microsoft 365 admin center under Copilot > Billing & usage and needs the Billing Administrator, AI Administrator, or Global Administrator role, plus an Azure subscription in the same tenant with a resource group and Owner or Contributor rights (both can be created during setup). First, on the Billing policies tab, create a billing policy: choose the Azure subscription and resource group, scope it to all users or specific users and groups, and optionally set a budget with alert recipients and thresholds. Second, on the Pay-as-you-go services tab, connect the policy to a service such as Microsoft Copilot Chat or SharePoint agents. The policy does nothing until it's connected. A tenant can have up to 50 billing policies.

Budgets send email notifications as spending approaches or passes the limit, but they don't enforce it: usage continues after the budget is exceeded. To stop pay-as-you-go usage, disconnect the billing policy from the service, and users lose access within about two hours. Deleting a billing policy removes the billing setup and disconnects its services. Before connecting a user-specific policy, disconnect any existing all-users policy.

Monitor spending on the Cost Management page for each pay-as-you-go service in the Microsoft 365 admin center, and in Microsoft Cost Management in the Azure portal, where you can analyze costs, set Azure budgets and alerts, and group by resource group or tags. Watch for patterns: steady daily use suggests moving a group to monthly licenses, while unexpected spikes may mean a policy scope is too broad.

The Microsoft Copilot usage report (Reports > Usage > Microsoft Copilot > Copilot) shows enabled users (licensed), active users, the active users rate, total and average prompts submitted, and adoption by app, over 7, 28, 90, or 180 days. A user counts as active only after an intentional AI action such as submitting a prompt, not just opening the Copilot pane. Data typically appears within 48 hours, and the user table includes everyone licensed in the past 180 days. Don't build official usage metrics from the Purview audit log, which is designed for security and compliance.

Copilot Analytics is the measurement and reporting pillar of Copilot controls. Operational reports in the Microsoft 365 admin center cover licenses, agents, and deployments. The Copilot Dashboard in Viva Insights gives strategic insights across readiness, adoption, impact (such as Copilot-assisted hours), and sentiment. Any business or enterprise Microsoft 365 customer with Exchange Online can view it without a paid Viva Insights or Copilot license, but at least 50 Copilot or Viva Insights licenses unlock agent insights, benchmarks, and group-level views. Its data covers the previous 28 days with up to a six-day delay. Customizable reports in Viva Insights can add organizational metrics from systems such as SAP, Salesforce, or Workday to measure return on investment.

Use the data to act: compare enabled and active users to target training or reclaim licenses, follow adoption over time rather than single snapshots, and look for outliers such as unusually heavy usage from one account. Privacy settings such as minimum group size and exclusion lists for the dashboard are managed by admins.

Users manage personal prompts in Copilot Chat. The bookmark icon next to a sent prompt saves it under Your prompts in Prompt Lab; saved prompts can't be edited in place, so users run or copy a prompt, change it, save it as new, and delete the original. Share prompt copies a direct link or shares with a team, whose members see it under Team prompts. Scheduling a prompt, available with a Microsoft 365 Copilot license and grounded in work data, runs it automatically at a set time and frequency without the user needing to be online.

Admins publish organizational prompts in the Microsoft 365 admin center under Copilot > Prompts, as an AI Administrator or Search Editor (currently a Frontier preview feature). They can create prompts one at a time or import up to 100 per CSV, publish up to 1,000 per tenant, pin up to four as suggested prompts, edit, export, and delete them. Published prompts appear in Prompt Lab, as suggestions in Copilot Chat, and as autosuggestions while typing, usually within about three hours. Deletion is permanent, so export first, and per-prompt analytics show active users and submissions over 7, 14, or 28 days.

Scheduled prompts are governed through policy. Setting the Allow the use of connected experiences in Office cloud policy to Disabled removes scheduled prompts for everyone, along with other connected experiences; previously scheduled prompts keep running until their schedules finish. Admins with the Power Platform Administrator role can inventory scheduled prompts with PowerShell.

In the Microsoft 365 admin center, Copilot > Settings > Data access > Agents sets who can access agents and which types of agents users can install, and links to the full agent inventory. For each agent, admins can enable, disable, assign it to specific users or groups, block, or remove it; users can only access agents the admin allows. A separate sharing control decides who can share agents with the whole organization: All users (the default), specific users or groups, or no users. Changing it affects only new sharing actions.

Using agents depends on license and agent type. Copilot Chat includes agents grounded in instructions and public websites; agents that use work data need a Microsoft 365 Copilot license or pay-as-you-go billing. Agents respect each user's permissions, so an agent never exposes content a user can't already open.

Building agents has its own requirements. Agent Builder and Copilot Studio are included for users with a Microsoft 365 Copilot license. Otherwise, Copilot Studio needs a tenant license plus a Copilot Studio User License for each maker, while people who only use a published agent need no special license and guests can't use Copilot Studio. Copilot Studio agents live in Power Platform environments, where roles assigned in the Power Platform admin center decide what makers can do: Environment Maker creates and edits agents, Environment Admin manages the environment, and Basic User only uses shared agents. Assign roles to security groups, separate development from production environments, and review access regularly.

In Agent Builder, a user describes the agent in natural language or fills in the Configure tab: name, description, instructions, knowledge sources such as SharePoint sites, folders, and files, web content, and Copilot connectors, optional capabilities such as code interpreter and image generation, and starter prompts. Templates provide preconfigured starting points. The user tests the agent, then selects Create. The new agent is private to its creator until shared.

Sharing uses two roles. Can edit makes someone an owner with full rights to edit, share, and manage the agent; Can chat lets them use it only. Groups can only be chat users, so owners must be added individually. When sharing, the owner can also share the agent's SharePoint files and folders if they have permission; otherwise the agent respects each user's access and omits content they can't open. Org-wide sharing makes the agent discoverable in the Agent Store, if the admin allows it.

SharePoint agents are created with the agent tool built into SharePoint by users with Edit permission or higher. Start from the site home page (+New > Agent), a library's command bar, selected files, or the agent chat pane. The agent is scoped to the selection, can add more sites, libraries, folders, or files as sources, and has a welcome message and up to three starter prompts. Agents created from a site home page are stored as .agent files in Site Assets > Copilots. A site's automatically created ready-made agent can't be edited.

A creator who wants an agent broadly available submits it to the org catalog. An admin reviews the request in the Microsoft 365 admin center (Agents > All agents > Requests, also reachable from the Agent overview's pending requests card) and approves it to publish in the Built by your org section of the Agent Store, or rejects it. Only AI Administrator or Global Administrator can approve requests; prefer AI Administrator. The creator's shared version and the published Agent Store version are managed separately.

A sound process has three stages: submission, with the agent's purpose, audience, and requested permissions; review, where admins involve business owners for context and security and compliance officers for data access risk; and decision, to approve, reject, or send back for changes. Document decisions, and review updates to approved agents again, especially when an update requests broader access.

For agents built in Copilot Studio, governance sits mainly in the Power Platform admin center. Data policies can block publishing to specific channels, restrict connectors, knowledge sources, and triggers, and prevent unauthenticated access; sharing limits keep uncertified agents from spreading; and application lifecycle management moves agents through development, test, and production environments.

In the Microsoft 365 admin center, Agents > Overview gives a 30-day snapshot: the number of agents in the registry, active users, agent run-time, and action cards for pending requests, agents without owners, agents at risk (high-severity risks from Entra, Defender, and Purview), and agents with exceptions. Some metrics, such as active users and agent run-time, start counting when the organization activates Agent 365 licenses. Only AI Administrator or Global Administrator can take governance actions such as approving requests or assigning owners.

The Agent Registry (Agents > All agents > Registry) lists every agent available to the organization, grouped as Microsoft agents, external partner-built agents, agents published by your org, and agents shared by creators, with filters for status, publisher type, channel, platform, and data source. Admins can export the list, upload a custom agent, and pin agents for users. Shared agents become ownerless when their creator's account is deleted; the Agents without owners view helps admins assign an owner, block, or delete them. Across the lifecycle of creation, approval, deployment, maintenance, and retirement, blocking stops users from using an agent while it stays in the inventory, and removing takes it out of the organization's inventory.

Usage data also appears in the Copilot usage reports under Reports > Usage and in the Copilot Dashboard. For agents built in Copilot Studio, the Power Platform admin center shows agent usage and adoption trends, lets admins set pay-as-you-go Copilot Credit caps and manage capacity, and applies data policies and environment controls; makers see per-agent performance on the Analytics tab in Copilot Studio.