Suppose you make a super-intelligent AI and run it on a computer. The computer has NO conventional means of output (no connections to other computers, no screen, etc). Might it still be able to get out / cause harm?

If humans were aware of it, probably many would try to break in from the outside.