What do “frontier AI labs” actually train their models on? Do we know? Do they even know? There are arguments to be made we should all know, and in detail.