It's going to be interesting to see how the circle is squared regards defending against autonomous weaponry. There's probably a large design space of offensive autonomous weapons that could only effectively be countered by a system with no human in the control loop. If the design / use of autonomous defensive systems is also prohibited, that would give an overwhelming advantage to any aggressor willing to defy any such a ban on offensive systems. If they're not banned, then the difference between a weapon designed to auto-target other weapons and one designed to auto-target humans is so blurry as to be virtually indistinguishable in terms of verifiable arms control.