Understand generative AI fundamentals
Free Microsoft Certified: AI Business Professional practice — 6 questions on Understand generative AI fundamentals, with explanations. No sign-up. Full 12-question mixed test →
An employee opens Copilot in Word and asks: "Summarize the customer contract stored in our SharePoint site, and check the latest public news about the client's parent company." Copilot returns one response that cites both the SharePoint document and a recent news article. Which statement correctly describes how Copilot handled the data protection boundary in this response?
Microsoft 365 Copilot grounds work-related answers in Microsoft Graph data (documents, mail, chats the user has permission to) under the enterprise's commercial data protection commitments, while web content is fetched separately via web search and is not covered by the same enterprise data boundary. The two remain distinct sources even when blended into one answer.
A finance manager wants a solution that checks an approvals mailbox every morning, extracts invoice data automatically, and posts a summary to a Teams channel with no one needing to type a prompt each day. Which type of Copilot experience best fits this requirement?
Agent experiences differ from chat experiences in that agents can be configured to run on a schedule or trigger and take action without a human initiating each turn, which matches the requirement for an unattended daily workflow.
Two employees type the exact same prompt into Microsoft 365 Copilot at the same time, grounded in identical documents, yet receive answers with noticeably different wording. What best explains this?
LLMs predict the next token based on probability distributions rather than retrieving a single fixed answer, so even with the same prompt and the same grounding data, the generated wording can differ between runs. This is a core, non-technical concept the exam expects business users to understand about why outputs vary.
A user asks Copilot Chat: "What was our exact revenue figure last quarter, based on the internal finance report?" Copilot responds with a specific confident-sounding number but provides no citation or link back to a source document. What is the most appropriate next step for the user?
Generative AI can hallucinate or fabricate specific-sounding details, and a lack of a citation is a signal that the figure may not be reliably grounded. Best practice, and a core exam concept, is that users must verify important facts against the authoritative source rather than trust confident-sounding output at face value.
An employee discovers that Copilot surfaced content from a sensitive HR file they technically had view permission to but were unaware existed, surprising their manager who believed that file was restricted. Which Responsible AI consideration does this scenario primarily illustrate?
Copilot only surfaces content the requesting user already has permission to view, but this scenario shows how existing overly broad or forgotten permissions can lead to unintended access and oversharing risk — a Privacy and Security concern that organizations must address through proper permission hygiene before deploying Copilot.
A team lead wants payroll tax withholding calculated to the cent for 500 employees, guaranteed to match authoritative published tax tables exactly for regulatory compliance. Which approach reflects the correct understanding of generative AI's limits?
Generative AI is a probabilistic text-generation technology and is not designed to guarantee exact deterministic numeric compliance calculations; regulated, to-the-cent computations should be handled by purpose-built deterministic systems, with generative AI potentially assisting with summarization or drafting around that output, not the calculation itself.
Ready for the real thing?
The full course: two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed explanations.
$99.99 $34.99 with code FREETEST33 — valid through September 19.
Get my $34.99 deal →