When you make messages like this, particularly if there is any threat involved, seem to me that only safe way is to have someone else, or even multiple people write it, and a different person deliver message.

I think that AI can use word pattern recognition to discover who is speaking, if they also are ones that wrote message.