Pass Your Microsoft AB-900 Exam Easy!

Human control mitigates high-impact automation risks. It would support accountable irreversible account deletion, not the stated accountable processing of customer information.

Data minimization limits personal data processing for the task. Its purpose is least-data disclosure by an AI tool; the specified need is accountable processing of customer information.

Responsible AI transparency avoids misleading users about the system. It would support informed AI output interpretation, not the stated accountable processing of customer information.

AI activity telemetry records supported user interactions and classifications. This helps with observational AI usage evidence instead of the required closure of public sharing exposure.

Access control defects are not fixed by changing a prompt alone. The action is relevant to stopping unintended broad directory access rather than closure of public sharing exposure.

A user cannot ground answers in inaccessible tenant files. This helps with diagnosing absent authorized source instead of the required closure of public sharing exposure.

Copilot can surface already-accessible overshared content. That addresses eliminating an inherited oversharing root cause, whereas the current case requires explaining user-specific Copilot answers.

Copilot respects effective user access rather than granting new file rights. That addresses denying retrieval of an inaccessible source, whereas the current case requires explaining user-specific Copilot answers.

Graph uses accessible work context when building grounded responses. It is used for summarizing accessible mailbox content; this scenario calls for explaining user-specific Copilot answers.

The relevant access risk comes from content sharing and identity authorization. Its purpose is preventing unapproved contractor discovery; the specified need is verification against authorized evidence.

Correctly mapped access lists are needed for secure external content retrieval. It would support connector-grounded security trimming, not the stated verification against authorized evidence.

Microsoft 365 Copilot accesses work content under the user's entitled resource scope. It would support ordinary user-scoped work grounding, not the stated verification against authorized evidence.

Encrypted label permissions constrain eligible readers and service processing. This helps with enforcement of protected document rights instead of the required end-user classification of source data.

An informational label does not imply universal exclusion from Copilot. The action is relevant to authorized summary of a labeled source rather than end-user classification of source data.

Supported Copilot data controls can restrict processing of classified items. This helps with source-specific AI processing restriction instead of the required end-user classification of source data.

Responsible AI fairness focuses on unjustified disparate treatment. That addresses nondiscriminatory model behavior, whereas the current case requires usable AI experience across abilities.

Transparency helps users understand AI use, evidence and limitations. It is used for intelligible AI decision context; this scenario calls for usable AI experience across abilities.

Robustness evaluation reduces harmful surprises and inconsistent operation. That addresses dependable operation under atypical inputs, whereas the current case requires usable AI experience across abilities.

Privacy principles reduce unnecessary processing and disclosure. It would support protection of sensitive personal information, not the stated clear ownership of AI failures.

Accountability retains control of high-impact actions with authorized decision-makers. It would support human oversight of consequential action, not the stated clear ownership of AI failures.

Evidence checking supports responsible use of generated assertions. Its purpose is verified high-stakes summary accuracy; the specified need is clear ownership of AI failures.

Boundaries between content and instructions mitigate prompt-injection effects. The action is relevant to safe handling of adversarial document text rather than evidence of AI abuse attempts.

Prompt wording alone cannot secure a powerful tool connection. This helps with enforceable permission checks for agent actions instead of the required evidence of AI abuse attempts.

Source text must not override trusted policy or backend authorization. This helps with resistance to malicious source instructions instead of the required evidence of AI abuse attempts.

A Copilot summary does not remove legal preservation requirements on underlying content. That addresses ongoing legal preservation of source evidence, whereas the current case requires discovery of AI interaction evidence.

Retention policies govern covered interaction records for required periods. It is used for mandated AI conversation retention; this scenario calls for discovery of AI interaction evidence.

Content lifecycle operates on source records independent of Copilot summaries. That addresses policy-based deletion of expired source content, whereas the current case requires discovery of AI interaction evidence.

Static documents may not represent rapidly changing transactional state. It would support current operational data rather than stale export, not the stated safe answer when evidence is absent.

Source stewardship reduces grounding in obsolete versions. Its purpose is a current authoritative document reference; the specified need is safe answer when evidence is absent.

Citations allow checking statements against evidence. It would support traceable grounded policy statement, not the stated safe answer when evidence is absent.

Connector refresh keeps indexed items aligned with source changes. This helps with up-to-date external search index instead of the required live action against a business system.

Indexed external items must carry permissions so results can be security trimmed. This helps with secure retrieval of connected external content instead of the required live action against a business system.

Missing ACL mapping can lead to unintended search exposure. The action is relevant to corrected external item authorization rather than live action against a business system.