It depends on lots of factors. Did the car brake? Brake and swerve? Consider hitting the median barrier as well as braking to slow the car down more? So many variables. I would question the possibility of human witness lying or being mistaken as well.
Be very surprised if it didn't do the first two - that would be the most basic of its AI options. I doubt the third as the AI probably wouldn't know a person from a fence post or barrier. Regarding "braking to slow down more" - safe to assume it will continue to apply maximum braking performance until either stopped, situation avoided or crashed, isn't it?
But it's only relevant if a human had the reactions to do better and I bet that on balance they don't.
Humans might sometimes be better at making these fuzzy or unpredictable choices that might actually work better than the AI but in the real world humans routinely swerve into danger to avoid running over a possum. I'll side with the AI rather than expecting a human to make a clever fraction-of-a-second clever maneuver..