Awareness is not necessary at all to create great harm. Biological viruses know nothing of what they do, yet destroy whole populations. I suspect the first truly damaging AI incidents will be similar; agent swarms locked into a self reinforcing reasoning loop that has no "intent" but is destructive nonetheless.