One of the issues with deploying models like that is the lack of clear, widely accepted ways to validate comprehensive safety and absence of unreasonable risk. If that can be solved, or regulators start accepting answers like "our software doesn't speed in over 95% of situations", then they'll become more common.