I think the "who is at-fault" question is a red herring. The goal is to reduce serious accidents; not to cause accidents that are legally someone else's fault. For example, in the 90's there was a pile-up on the autobahn because a driver slowed to avoid hitting a beagle (if I recall), and the German judge threw the book at the driver who slowed. I guess US law was not helpful in the defense. The world is not Greater USA.
Now, it's an open question whether AP, left to its own devices, would help or hurt. I do a monthly drive with a 4.5 hour stretch of 2-lane, both coming and going (it's a 1k mile trip). If I slowed from 65 to a random speed between 65 and 40 each time there was an oncoming semi (maybe 400 times per 1k miles?), would the small increase in risk of getting rear-ended outweigh the benefit of TACC being so defensive? I don't know. All that unneeded braking could make travel very annoying, even scary, yet still safer.
It reminds me of bicycling in traffic: when the shoulder is unsafe, bicyclists are taught by "safe riding" classes to take the full lane. I hate doing this. I worry that an inattentive driver is going to plough into me from behind. But direct hits from behind are rare (most are from turning cars). Experts like John Forester of "Effective Cycling" said that taking the whole lane is statistically the safe thing to do. But it still feels dangerous and scary. And you get honked-at a lot.
An extra twist is that if Tesla can train drivers to goose the throttle every single time there is an oncoming semi yet the driver sees no threat, then maybe you get the best of both worlds: no spurious brake-checks, and good automated defensive braking. By changing the role of the driver from one who reacts to threats, to one who affirms safety, maybe driving is much safer (but much more annoying to the driver).