One problem with this reasoning is that the researchers really didn't know what they were doing with 100% certainty. Their code could have accidentally affected the stability control subsystem that most cars have nowadays -- the one that's designed to apply full braking to a single wheel to recover from a skid. In fact, just corrupting the data from the steering-wheel angle sensor could have had that effect (which I personally find rather terrifying in itself.) Good job, guys, now you've caused a 70 MPH rollover in traffic.

The right way to do this would have been for the researchers to call the police up front and arrange a demonstration on a closed road with police escort. That would have lent the video more credibility and shielded the researchers from liability, while addressing any concerns about safety or ethics.

Firing a gun at someone is dangerous. A car slowing down and/or with reduced visibility on a busy highway is dangerous. Then either of these situations happen by accident, we understand that there's not a lot to do about them because there was no intentional behavior that needs correction. When they are purposefully done, that's endangering people, and is unacceptable behavior that needs correcting. In this respect, they are no different.