The investment control software of large financial companies seems the most likely source of rogue AI.

Charlie Stross, in his 2005 novel Accelerando, implicitly suggested "financial device" AI as the most likely seed for rogue AI. Last year, David Brin independently and explicitly promoted the possibility. The idea seems more likely today.

With a multi-million dollar salary from an evil bank, you can contribute to species survival.
You're welcome. And I'm dead serious.

Apparently, Brin first wrote a full outline of the idea here: http://ieet.org/index.php/IEET/more/brin20111217 More recently: http://www.newscientist.com/article/mg21528735.700-artificial-intelligence-to-sniff-out-bankers-scams.html The New Scientist link is from Brin's (excellent) blog, http://davidbrin.blogspot.com/2012/07/bulletins-from-transparency-front.html?showComment=1342734425933#c5447006304917530539

And this post from the ever useful DailyKos makes several suggestions: http://www.dailykos.com/story/2011/06/12/981390/-Sci-Fi-Fantasy-Club:Magic-and-Fantasy?via=siderecent

L. E. Modesitt has written consistently rational fiction, and improved steadily as a writer. He scores on both counts. http://en.wikipedia.org/wiki/L._E._Modesitt