I'm not really convinced the key difference between "bot that notifies people they've leaked their CC" and "bot that re-shares leaked CC pictures" is 'punch up

I'm not really convinced the key difference between "bot that notifies people they've leaked their CC" and "bot that re-shares leaked CC pictures" is 'punch up not down'. -- One is trying to prevent people losing stuff, the other is trying to encourage stealing. I don't think "punch up not down" is morally equivalent to "be helpful, not malicious". With RLHF'd LLM chatbots, the output tries to balance being accurate, being helpful, and being inoffensive. -- I'd worry that "punch up not down" mandate dictates it's better for a bot to be inaccurate rather than to say something offensive. -- I think the "punch up not down" people don't understand that they might not be the ones who decide what gets considered 'offensive' or not.