Here is my example, now on YouTube. In this case Sophy doesn't lift off, its speed just changes unnaturally.
I could be TOTALLY wrong, but this all looks like two systems competing. I see two systems at work in the video above. I see AI and I see GAME rules.
Just by the video evidence, and what we see in game. it appears that there is AI to control the car (as in pilot it around the track), while at the same time, there are game rules governing the pace of the car (the infamous rubber band effect). We can see from the in car view that the AI is not lifting, but the car slows none the less. The AI didn't make a mistake; a game rule was triggered.
Back in GT5, when the AI difficulty was unlocked, increasing the difficulty didn't make the AI better, it changed the field of cars so the AI had faster cars. Even still, in the custom races, the AI never changes, or changes very little. Switching the AI from Beginner to Professional does very little, but changing the cars DOES increase the overall difficulty. Presumably because the difficulty of the event is determined by the game rules, which dictate the gap between you and the other cars.
Could it be that for nearly 30 years, we've been complaining about "The AI" to the point that Sony invested in Sophy AI, but the real problem has been core game rules, controlling the pace of cars, and it has nothing to do with AI behavior at all? Because we have all noticed that the "AI" is so much better in the hot pepper races where the pace is allowed to be more competitive.
I'd bet you a dozen doughnut that this is where the "AI" problem resides