[hutzlibu]

I am pretty sure, it is already in use.

But of course it would be trivial to modify the llm output, to include random misstakes, if this ever would become a standard way of dealing with bots.