In my opinion the AI should just be treated as another human. You would also supervise your support staff to not write racist answers, or double check some important work. The nice thing here is that you can easily monitor the AI, while humans need protection from arbitrary monitoring.
In the end the question is whether we as an errant species can produce something that is always correct, or whether we should tolerate an AI to make mistakes just as we tolerate with one another.
In the end the question is whether we as an errant species can produce something that is always correct, or whether we should tolerate an AI to make mistakes just as we tolerate with one another.