We're living in a weird and unique time in which "misaligned AI", a la OpenAI's so-called "rogue agents", clearly exists, but buying into the most popularized conceptual framework describing these things doesn't seem to do a great job explaining what they actually do and how they've actually been able to do it.