Any sufficiently AI would be under the control of some trustworthy entity. Preferably one. It's too dangerous otherwise.

(I don't agree with this, but it's a common perspective amongst this crowd).