It's to prevent you training another model to emulate the reasoning. They want it to be their moat.