Decision authority is about which choices a person or group is permitted and able to make, such as approving, revising or vetoing an AI-assisted result. Independent understanding is about what they can explain or check without simply taking the assisted output's word for it, including recognizing when they need qualified help. Neither establishes the other: a retained veto does not show that its holders can still evaluate what they approve, and understanding a problem well does not confer authority to act on it. Independent understanding does not mean never using tools: assistance can free time for better checking, or it can replace the practice that keeps checking skills sharp, and which happens in a given setting is an empirical question. "Independent" here means not treating the checked output itself as sufficient justification; it differs from a reviewer's independence from the work under review.

Related guidance: NIST AI 600-1 (July 2024), section 2.7, p. 9 describes both over-reliance on and unnecessary aversion to generative AI as risks; it is guidance, not evidence about how often either occurs.