Human review of the AI's determination
If you think about it, the human reviewer is not likely to have access to any additional information and is not likely to have access to any additional decision-making criteria (even moral and ethical policies). Any additional information and criteria that might become available would immediately (*) be incorporated into the AI system.
THIS-> Effectively the human reviewer is left to nod their head, and press 'PROCEED'.
* Exception: The humans might be quicker to 'reprogram' with a memo or direct order on any given day, whereas the software update to the AI might need weeks or months.