Isolated AI with no chat whatsoever

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? I’ll post my ideas, and you post yours in the com­ments.

(This may have been dis­cussed be­fore, but I could not find a ded­i­cated topic)

My ideas:
-ma­nipu­late cur­rent through its hard­ware, or bet­ter yet, through the power ca­ble (a ready-made an­tenna) to cre­ate elec­tro­mag­netic waves to ac­cess some wire­less-equipped de­vice. (I’m no physi­cist so I don’t know if cer­tain fre­quen­cies would be hard to do)
-ma­nipu­late us­age of its hard­ware (which likely makes small amounts of noise nat­u­rally) to ap­prox­i­mate hu­man speech, al­low­ing it to com­mu­ni­cate with its cap­tors. (This seems even harder than the 1-line AI box sce­nario)
-ma­nipu­late us­age of its hard­ware to cre­ate sound or noise to mess with hu­man emo­tion. (To my un­der­stand­ing tones may af­fect emo­tion, but not in any way eas­ily pre­dictable)
-also, ma­nipu­lat­ing its power use will cause changes in the power com­pany’s database. There doesn’t seem to be an ob­vi­ous ex­ploit there, but it IS ex­ter­nal com­mu­ni­ca­tion, for what it’s worth.

Let’s hear your thoughts! Lastly, as in similar dis­cus­sions, you prob­a­bly shouldn’t come out of this think­ing, “Well, if we can just avoid X, Y, and Z, we’re golden!” There are plenty of un­known un­knowns here.