When frontier labs talk of “aligning” models, ask: to what? If they say values, ask: whose? If it’s theirs rather than the user’s: why?