Delegating to a committee was moral cover. Delegating to an AI is cleaner.
The model can't leak. It can't refuse. You reframe the prompt until it agrees, then say the AI decided.
Agency laundering, perfected.
https://www.joanwestenberg.com/p/the-art-of-agency-laundering
@Daojoan
I've frequently made the case that I don't care whether responsibility is assigned to the AI or maintained with the user. But it has to be one or the other.
If the users of an AI model pass the buck to the model when something horrible happens, fine - then lock up the model in an AI jail for a suitable number of years where no one can use it.