I personally think that frontier LLMs are already at the point where I trust them to be a better judge of human behavior than humans overall.

Absolutely. But they still hallucinate, I wouldn't want them to take action themselves, just flag things for human review.