You don’t know what you’re talking about.

It is not as dumb as just looking for spikes in acceleration, but specific patterns and integrating acceleration to get position and velocity history in order to match situations that look like car crashes. Obviously the real world is demonstrating some bugs with bad matches that didn’t get found in testing, but this isn’t surprising or some big failure. Unfortunate it is bothering emergency workers, but is it actually impacting emergency services?