Ah, I was missing the obvious. Thanks for pointing that out. I thought that the scandal was that these instructions applied to Meta's AI; if they were reaching users' own AIs that's a whole different story.