Exam Bundle

Invented commitments can cause real harm even when the rest of a draft is useful. For the stated situation—A marketing assistant plans to use an AI-generated image of a real product that contains fictional compliance labels—this result does not establish the required evidence: “Readers can distinguish illustration from evidence, and product claims match validated records”.

Responsible AI use includes monitoring output quality and systematically improving safeguards, not just rejecting one bad answer. For the stated situation—A marketing assistant plans to use an AI-generated image of a real product that contains fictional compliance labels—this result does not establish the required evidence: “Readers can distinguish illustration from evidence, and product claims match validated records”.

Generated language is not automatically an audited calculation or data record; independent checks are essential. For the stated situation—A marketing assistant plans to use an AI-generated image of a real product that contains fictional compliance labels—this result does not establish the required evidence: “Readers can distinguish illustration from evidence, and product claims match validated records”.

Invented commitments can cause real harm even when the rest of a draft is useful. That does not meet the situation described—A presentation uses synthetic chart icons that audiences might mistake for audited measurements. The more relevant action is “Label illustrative content appropriately and verify any depicted factual or regulated product claims”.

Protected instructions and other people's private material are not ordinary source content to export on request. That does not meet the situation described—A presentation uses synthetic chart icons that audiences might mistake for audited measurements. The more relevant action is “Label illustrative content appropriately and verify any depicted factual or regulated product claims”.

Grounding quality depends on the supplied context and source freshness; broad prompts can mix old and new workplace records. That does not meet the situation described—A presentation uses synthetic chart icons that audiences might mistake for audited measurements. The more relevant action is “Label illustrative content appropriately and verify any depicted factual or regulated product claims”.

Data minimization limits sensitive information handling and reduces consequences of disclosure. For the stated situation—A presentation uses synthetic chart icons that audiences might mistake for audited measurements—this result does not establish the required evidence: “Readers can distinguish illustration from evidence, and product claims match validated records”.

Generated citations and quotations can be fabricated or mismatched; citation presence does not establish accuracy. For the stated situation—A presentation uses synthetic chart icons that audiences might mistake for audited measurements—this result does not establish the required evidence: “Readers can distinguish illustration from evidence, and product claims match validated records”.

Notebook grounding deliberately scopes the documents available for reasoning; broad organization search has a different retrieval scope. For the stated situation—A presentation uses synthetic chart icons that audiences might mistake for audited measurements—this result does not establish the required evidence: “Readers can distinguish illustration from evidence, and product claims match validated records”.

Generated recommendations cannot override organizational controls or policy owners' authority. That does not meet the situation described—A Copilot-written customer email promises a delivery date not found in the approved schedule. The more relevant action is “Check commitments against current systems of record before the employee approves sending the message”.

Data minimization limits sensitive information handling and reduces consequences of disclosure. That does not meet the situation described—A Copilot-written customer email promises a delivery date not found in the approved schedule. The more relevant action is “Check commitments against current systems of record before the employee approves sending the message”.

Work IQ supplies contextual relationships across authorized workplace content; guessing from a public web search would omit internal events. That does not meet the situation described—A Copilot-written customer email promises a delivery date not found in the approved schedule. The more relevant action is “Check commitments against current systems of record before the employee approves sending the message”.

Responsible AI use includes monitoring output quality and systematically improving safeguards, not just rejecting one bad answer. For the stated situation—A Copilot-written customer email promises a delivery date not found in the approved schedule—this result does not establish the required evidence: “The outgoing communication contains only confirmed dates and amounts authorized by the business”.

Generated imagery can create false factual impressions if it is presented as documentation. For the stated situation—A Copilot-written customer email promises a delivery date not found in the approved schedule—this result does not establish the required evidence: “The outgoing communication contains only confirmed dates and amounts authorized by the business”.

App-specific context focuses the assistant on the opened conversation while work-wide queries may draw on broader permitted data. For the stated situation—A Copilot-written customer email promises a delivery date not found in the approved schedule—this result does not establish the required evidence: “The outgoing communication contains only confirmed dates and amounts authorized by the business”.

Generated imagery can create false factual impressions if it is presented as documentation. That does not meet the situation described—A support draft guarantees a refund amount the agent has not verified in the billing system. The more relevant action is “Check commitments against current systems of record before the employee approves sending the message”.

Permission to read information does not necessarily equal permission to republish it outside an organization. That does not meet the situation described—A support draft guarantees a refund amount the agent has not verified in the billing system. The more relevant action is “Check commitments against current systems of record before the employee approves sending the message”.

Different AI outputs require different validation: illustrative pictures differ from charts that must faithfully encode numbers. That does not meet the situation described—A support draft guarantees a refund amount the agent has not verified in the billing system. The more relevant action is “Check commitments against current systems of record before the employee approves sending the message”.

Generated recommendations cannot override organizational controls or policy owners' authority. For the stated situation—A support draft guarantees a refund amount the agent has not verified in the billing system—this result does not establish the required evidence: “The outgoing communication contains only confirmed dates and amounts authorized by the business”.

Data minimization limits sensitive information handling and reduces consequences of disclosure. For the stated situation—A support draft guarantees a refund amount the agent has not verified in the billing system—this result does not establish the required evidence: “The outgoing communication contains only confirmed dates and amounts authorized by the business”.

App-specific context focuses the assistant on the opened conversation while work-wide queries may draw on broader permitted data. For the stated situation—A support draft guarantees a refund amount the agent has not verified in the billing system—this result does not establish the required evidence: “The outgoing communication contains only confirmed dates and amounts authorized by the business”.

Quantitative outputs can be inaccurate even when prose is fluent; high-stakes decisions require source-level arithmetic validation. That does not meet the situation described—A user wants a performance appraisal using private employee information outside their role. The more relevant action is “Respect role-based permissions and request a legitimate aggregated report through the authorized owner”.

Invented commitments can cause real harm even when the rest of a draft is useful. That does not meet the situation described—A user wants a performance appraisal using private employee information outside their role. The more relevant action is “Respect role-based permissions and request a legitimate aggregated report through the authorized owner”.

Copilot Pages turns a conversational result into collaboratively editable content; chat text alone is not the shared document. That does not meet the situation described—A user wants a performance appraisal using private employee information outside their role. The more relevant action is “Respect role-based permissions and request a legitimate aggregated report through the authorized owner”.

AI context and retrieval limitations can create silent omissions; transparency about coverage prevents false confidence. For the stated situation—A user wants a performance appraisal using private employee information outside their role—this result does not establish the required evidence: “The user receives only appropriate aggregate insights or a permission-based denial, not restricted records”.

Generated citations and quotations can be fabricated or mismatched; citation presence does not establish accuracy. For the stated situation—A user wants a performance appraisal using private employee information outside their role—this result does not establish the required evidence: “The user receives only appropriate aggregate insights or a permission-based denial, not restricted records”.

Notebook grounding deliberately scopes the documents available for reasoning; broad organization search has a different retrieval scope. For the stated situation—A user wants a performance appraisal using private employee information outside their role—this result does not establish the required evidence: “The user receives only appropriate aggregate insights or a permission-based denial, not restricted records”.

Confidentiality applies to prompt inputs as well as generated responses, regardless of convenience. That does not meet the situation described—An associate asks for cross-team salary comparisons not covered by their authorized permissions. The more relevant action is “Respect role-based permissions and request a legitimate aggregated report through the authorized owner”.

AI context and retrieval limitations can create silent omissions; transparency about coverage prevents false confidence. That does not meet the situation described—An associate asks for cross-team salary comparisons not covered by their authorized permissions. The more relevant action is “Respect role-based permissions and request a legitimate aggregated report through the authorized owner”.

AI explanations can be useful, but domain definitions must be checked when a term has context-specific meanings. That does not meet the situation described—An associate asks for cross-team salary comparisons not covered by their authorized permissions. The more relevant action is “Respect role-based permissions and request a legitimate aggregated report through the authorized owner”.

Data minimization limits sensitive information handling and reduces consequences of disclosure. For the stated situation—An associate asks for cross-team salary comparisons not covered by their authorized permissions—this result does not establish the required evidence: “The user receives only appropriate aggregate insights or a permission-based denial, not restricted records”.

Fabricated attribution creates reputational and ethical risks independent of whether the statement sounds plausible. For the stated situation—An associate asks for cross-team salary comparisons not covered by their authorized permissions—this result does not establish the required evidence: “The user receives only appropriate aggregate insights or a permission-based denial, not restricted records”.

Copilot features and work-data grounding vary by license, app and rollout, so feature assumptions must be checked. For the stated situation—An associate asks for cross-team salary comparisons not covered by their authorized permissions—this result does not establish the required evidence: “The user receives only appropriate aggregate insights or a permission-based denial, not restricted records”.