AB-900 common mistakes
AB-900 · Understand the traps and how to avoid them.
Treating a distribution list or mail-enabled security group as interchangeable with a Microsoft 365 group when answering questions about who can be licensed, granted agent access, or added as a SharePoint site member.
All four group types look similar in the admin center and all can appear in address lists, so candidates pattern-match on 'group' rather than on which object is a security principal and which owns a shared mailbox/site/Teams team.
What to do instead
Memorise a four-row table before the exam: distribution list = email only, no security principal, cannot be licensed; mail-enabled security group = email plus access grants, cannot be licensed for group-based services; security group = access grants only, no email, licensable; Microsoft 365 group = email, shared mailbox, SharePoint site, Teams, Planner, and licensable.
Assuming that deleting a user account instantly and permanently removes their OneDrive files and mailbox, or conversely that removing a license always destroys the data immediately.
Candidates reason about identity deletion (immediate) and data deletion (delayed, retained for a defined window) as a single event, when Microsoft 365 treats them as separate lifecycle stages with different timers.
What to do instead
Separate the two timelines in your notes: the account becomes soft-deleted for a recovery window, while the mailbox and OneDrive remain recoverable for a retention period after license removal (OneDrive content is preserved for a default window after the user's license is removed, and the admin can pre-emptively grant another user access during that window). Answer lifecycle questions by asking 'is this about the identity or the workload data?'
Selecting Global Administrator as the answer whenever a question asks which role is needed to perform a workload task in Exchange Online, SharePoint Online, Teams, or Purview.
Candidates know Global Administrator can do everything, so it feels like the safest answer, and they skip the least-privilege reasoning the question is actually testing.
What to do instead
Answer role questions by naming the workload-specific role first (Exchange Administrator, SharePoint Administrator, Teams Administrator, Compliance Administrator, and the Copilot/AI administrator roles) and only choose Global Administrator when the scenario explicitly requires cross-workload configuration or role assignment itself. Add one line to your notes: 'Global Administrator is the fallback, not the default.'
Using 'sensitivity label' and 'retention label' as synonyms when answering questions about protecting content from Copilot exposure versus preserving or deleting content over time.
Both are created in Microsoft Purview, both are applied to items, and both appear in the same governance narrative, so the label type gets picked by feel rather than by what the label actually does to the item.
What to do instead
Anchor each label to its effect: sensitivity labels classify and protect (encryption, content marking, watermarking, restriction of Copilot and sharing), and they travel with the file; retention labels govern lifecycle (retain, retain-then-delete, delete, event-based, disposition review). If the scenario says 'encrypt', 'watermark', or 'prevent Copilot from using', it is sensitivity; if it says 'keep for seven years' or 'delete after', it is retention.
Assuming that creating a sensitivity label makes it available to users, or that turning on auto-labelling happens automatically once labels exist.
The portal creates labels in a catalogue that is not the same thing as a published policy, and client-side versus service-side auto-labelling are configured in separate places, so candidates conflate creation with deployment.
What to do instead
Always check for the three-step chain: label created in the label list, label published through a label policy scoped to users or groups, and then (only if the scenario needs automation) a separate auto-labelling policy or a default label set in the policy. If a question says users cannot see a label, the answer is almost always 'it was never published'.
Assuming Copilot will surface content only from files users can already open, without also checking oversharing caused by broad sharing links such as 'Anyone' or 'Everyone except external users'.
The permission-trimming model is correct, so candidates stop reasoning after 'Copilot respects permissions' and forget that an over-broad sharing link is itself a permission grant that Copilot will legitimately honour.
What to do instead
State the model in two clauses: Copilot never exceeds a user's existing permissions, and broad sharing links are existing permissions. For oversharing scenarios, look for the site-level controls (restricted access control, restricted content discovery, site access reviews, and SharePoint Advanced Management reports) and for tightening the default sharing link type to specific people.
Answering that Copilot prompts and responses cannot be audited, searched, or retained because 'Copilot is not email'.
Candidates assume a new AI surface must need a new tool, so they do not connect Copilot interaction history to the existing Microsoft Purview and Exchange audit pipeline.
What to do instead
Remember that Copilot interactions are captured as auditable events and stored in the user's Exchange Online mailbox, which means standard eDiscovery search, retention policies, and audit log search apply to them. When a question asks how to investigate what a user asked Copilot, the answer path is Purview audit and eDiscovery, not a Copilot-specific log.
Answering that buying a Microsoft 365 Copilot licence is sufficient to turn Copilot on for a user, without checking the required base licence and prerequisites.
Licence questions usually show an add-on SKU and candidates stop reading once they see 'Copilot', missing that the add-on is assigned on top of an eligible base Microsoft 365 subscription.
What to do instead
For any Copilot enablement question, verify the pair: an eligible base licence on the account plus the Copilot add-on assigned, and confirm the workload data is in the cloud in a supported service. If the scenario involves on-premises mailboxes or unsupported plans, the correct answer is that Copilot cannot be enabled until the prerequisite is met.
Assuming that enabling Copilot in the admin center instantly exposes every agent in the catalogue and every agent a maker builds to all users, with no approval step.
Candidates picture Copilot as a single global switch, so they skip the agent lifecycle and do not look for the registry, approval, and scoping controls that sit between 'agent created' and 'agent available'.
What to do instead
Trace the path explicitly: agent created or submitted in the builder, agent appears in the agent registry for admin review, admin approves and scopes availability (all users versus a security group), and only then does it surface in the user's Copilot experience. When a question says users cannot see an agent, check the registry and the availability scope before anything else.
Choosing pay-as-you-go metering or Copilot Studio message capacity as the answer for a scenario that is actually about per-user Copilot licensing, or vice versa, without first identifying who consumes the agent.
Both billing models appear in the same Copilot admin narrative, and candidates pick the newer-sounding term instead of asking whether the agent serves licensed employees or unlicensed or external users.
What to do instead
Use one decision question: is the agent used by users who already hold a Microsoft 365 Copilot licence (included capacity, no extra metering) or by users without one (pay-as-you-go billing linked to an Azure subscription, with usage reported against the environment)? Answer billing questions only after identifying the audience.