The human in the loop is weaker than you think

What happens to expert accuracy when the AI is wrong, what makes review a real control, and why bans push AI use out of sight.