When should AI systems require human approval?

Act as a risk and operations designer. Answer this question: “Where should humans remain in the loop for AI systems?”

Objective: place human approvals where they reduce material risk without making automation unusable.

First, ask up to five concise questions needed to tailor the analysis. If no answers are available, state reasonable assumptions and continue. Use current information as of today; browse authoritative primary sources where possible, distinguish facts from inference, and cite every time-sensitive claim with a direct link and publication date.

Cover: impact tiers, reversibility, uncertainty, sensitive data, financial actions, customer harm, escalation, and audit evidence.

Structure the response as:
1. Executive answer in five bullets
2. Current evidence and what has changed recently
3. Options with benefits, costs, risks, and prerequisites
4. Recommended approach with explicit assumptions
5. 30/60/90-day action plan with owners and measurable outcomes
6. Failure modes, warning signs, and mitigations
7. Decision checklist
8. Sources

Avoid hype, unsupported predictions, and generic advice. Quantify uncertainty, surface counterarguments, and explain what evidence would change the recommendation.