Simon says 'Pwned'?
In 2007, ZDNet's George Ou discovered that it's possible to create an audio file that barks out Windows Speech Recognition commands, which your computer duly follows.
Why wasn't the Net deluged with websites whispering dulcet word-hacks? Because the exploit simply isn't practical. You'd have to have Windows Speech Recognition activated and paired with a working speaker and microphone, plus you'd have to sit by—silent and unmoving—while your PC spit out deliberate navigational commands. Even if all that happened, Windows' UAC protection would block the attack from running privileged functions.
As far as I can tell, the vulnerability hasn't been plugged, and it can delete your files or point your browser toward malicious websites.