I took the question as-is (I think, misinterpretation is possible) and assumed there was such a thing as an ethical LLM when I voted.
But in reality, I don't think it's possible. It's as if someone discovered the Bubble Sort, saw that the algorithm blew up in large data sets, and decided the solution was special Bubble Sort hardware that consumed ten times as much power as the CPU it was attached to.
LLMs are the Bubble Sort of AI.