What I’ve found regarding this is, if the highlights are exposed at and above the sensor saturation point, the data stacks up there, as illustrated in the histogram. When white balance correction is applied, it spreads the three channels appropriately, but the stacked saturated data splits into three peaks, one for each channel. As you then move your data to do, say, positive exposure correction, you move each of those peaks to the display white bound, where they again start to stack up. If the green peak is leftmost, the highlight takes on a magenta cast because it’s green component is not at “white”.
If you increase the exposure correction enough, all three peaks eventually stack up at white and the magenta goes away. The difference between softwares is how aggressive is the automatic scaling for display.
I’ve been dealing with this specific thing in coming up with a batch workflow. For that, I want aggressive scaling so the proof image I’m making in batch looks nice. But if process ‘by hand’, I want to see it happen and deal with it manually.