Hello 
I was thinking about this too recently. It seems that we are on the same wavelength of AI testing 
In my case, I came up with a slightly workaround solution, if you can call it a solution. Since we assume that personal communication with AI may someday become public, then the Forum Administrator himself can start a private conversation (as I did) with AI, and then add a certain group of people to the conversation, having previously notified them that this conversation may become (possibly will definitely become) publicly available in the future. And to maintain relative confidentiality (if you can call it that) in some topics I use (still experimentally) hidden replies plugin + I use hiding tags on a message to hide some attributes (including usernames). Thus, after publishing a topic publicly, other users will only be able to see the AI’s answers (if it is an artist, then users will be able to see an additional expandable statement according to which the answer was given).
P/s. In the meantime, let me suggest you read this topic which I recently published , maybe she will be interesting to you 