I think, what is happening here is what the photofinishing industry calls “subject failure”. You might think this jargon misses the mark, but I think it’s spot on.
Our high-speed photofinishing printers analyzed each frame of a color negative or slide. The gathered data was used to set the intensity and color tint of the exposing light. I am talking color photo paper, which was subsequently developed to make a finished photographic print. This technique custom color balances each film frame. Without this degree of sophistication, photo labs were forced to re-print a high percentage of film frames. These techniques evolved, and I believe white balance algorithms are likely spin-offs.
Anyway, the photofinishing printer was plagued by “subject failure”. The scan picked up irrelevant background colors, which by computer logic were considered relevant. A baby on green grass or red blanket overwhelmed the logic and the printer attempts to correct (white balance). The irrelevant expanse induces an error and the result is an off-color result. Baby on green grass becomes too magenta, baby on red blanket too cyan; the unwanted cast is the complementary color (opposite).
Subject failure can be mitigated by high-powered software and/or human intervention. This is likely too much to ask of a camera at this juncture of camera evolution.