Which Responsible AI Principle Covers Human Oversight of AI Decisions?

Why the Answer Is Correct

Microsoft's six responsible AI principles are Fairness, Reliability and safety, Privacy and security, Inclusiveness, Transparency, and Accountability. The Accountability principle states that people, not the AI system, remain answerable for how AI operates and for the decisions it informs, which is exactly what the sentence in the exhibit describes: human reviewers oversee AI-generated decisions and carry ultimate responsibility. A community member (buianhthang) reached the same conclusion, writing that "people must remain accountable and responsible for how AI systems operate." Governance mechanisms such as human-in-the-loop review boards, sign-off workflows, and escalation paths exist precisely to make that accountability real. Because the sentence pairs human oversight with ultimate ownership of the final output, Accountability is the only principle that completes it without contradicting the wording.

Why Other Candidate Answers Are Wrong

Transparency is the most tempting distractor, but it requires only that AI systems be understandable and that their capabilities and limitations be communicated — it does not assign responsibility to a person. Fairness concerns equitable treatment across groups, and nothing in the sentence mentions bias, demographics, or disparate impact. Reliability and safety addresses consistent, dependable, and harm-free operation, which describes system behavior rather than human ownership of decisions. Privacy and security governs data handling and protection, and Inclusiveness is about empowering and engaging all people; neither maps onto the phrase about reviewers taking responsibility for the final output.

Community Comment Notes

Only one substantive comment is recorded here, and it votes Accountability with reasoning that aligns with Microsoft's published principle, so it does not conflict with the verdict on this page. The comment's key phrase, "take ultimate responsibility for the final output", is the same signal a test taker should key on in the exhibit image. Because the hotspot image is the only place the exact sentence appears, the safest technique is to read the sentence stem for the ownership verb (reviewers, responsible, answerable) before scanning the principle list. No commenter raised a competing principle, so community consensus and the official principle definition agree in this case.