One trust breaking issue is we still can't know why the LLM makes specific choices.

Sure we can ask it why it did something but any reason it gives is just something generated to sound plausible.