If I trust those numbers, and its not easy to trust anything from Tesla, then I could conclude that Teslas with autopilot are safer per mile on the easy stretches of highway where people are more comfortable enabling it than everyone is anywhere. It would be amazing if they couldn't match that extremely low bar, more comparable statistics would be comparing with other vehicles when their cruise controls are enabled.
With Tesla being constantly under the microscope, it would be difficult to not trust the data. They know that one misstep will seal their doom.