You want a plausible scenario for destruction via AI? Here's one. Some idiot governments put a neural network behind missile defense or nukes. Because ANNs are

You want a plausible scenario for destruction via AI? Here's one. Some idiot governments put a neural network behind missile defense or nukes. Because ANNs are uninterpretable black boxes, at some point the system does a counterdefense against nothing or grossly overreacts to some stimulus. Tada, absolute destruction and chaos. This might seem far fetched, but is it really that out-there of a scenario?