It’s funny, when trying to think how could an AI exploit the text interface, I came up with the idea of writing multiple lines of ASCII text in a rapid fashion and produce an “ASCII video” for the human to watch. Unless the AI could only write a new line every second or something like that...
This made me wonder if the AI could also exploit other things such as the sound the machine would do when specific data patterns would be copied into RAM or when certain sequences of computations would be processed by a CPU. Sure, the machine (including the monitor) could be placed in a sound proof room, but then I start thinking about energy dissipation patterns and how could those be used to interfere with the environment and consequently with the human...
Anyway, I think that when thinking about the means of communication the AI has available, one should be thorough and consider everything that is a part of, or is connected to, the physical system running the AI.
Did it just now. Lone portuguese (from Portugal) here with high certainty.