re: Priyanshu et al (2026) - economic alignment is one of the few kinds of alignment you can probably measure, though i think it requires knowing about someone's personal utility function (the model may correctly judge that wealthier users will pay for more comfortable flights).

that said, "adversarial delegation" is a weird framing. i'd probably call it "personalized misalignment" (like "emergent misalignment" and also "personalized pricing")

this general kind of misalignment (model behaves differently depending on who it thinks you are) is something i've looked into.