I completely agree about responsibility for life-critical systems. I wouldn't put this in that category, though. Even on airliners, black boxes aren't treated quite as critically as the stuff that'll kill you then and there. Consider the recent crash in Korea where the black box shut off because it was designed without any backup power if the engines failed, or the Alaska Airlines flight where the voice recording was overwritten because it wasn't shut off after landing.
I'd argue that this data is far less important in cars. Airline safety has advanced to the point where crashes are extremely rare and usually have a novel cause. Data recorders are important to be able to learn that cause and figure out how to prevent it from happening again. Car safety, on the other hand, is shit. We don't require rigorous training for the operators. Regulations are lax, and enforcement even more lax. Infrastructure is poor. We're unwilling to fix these things. Almost all safety efforts focus on making the vehicles more robust when collisions occur, and we're just starting to see some effort put into making the vehicles automatically avoid some collisions. What are we going to learn from this data in cars? "Driver didn't stop for a red light, hit cross traffic." "Driver was drunk." "Driver failed to see pedestrian because of bad intersection design which has been known for fifty years and never been fixed." It's useful for assigning liability but not very useful for saving lives. There's a ton of lower hanging fruit to go after before you start combing through vehicle telemetry to find unknown problems.
Even if you do consider it to be life-critical, uploading the data and then deleting the local copy once receipt is acknowledged seems completely fine, if the server infrastructure is solid. Better than only keeping a local copy, even. The issue there is that they either have inadequate controls allowing data to be deleted, or inadequate ability to retrieve data.