When AI Holds the Red Button, Deterrence Could Put Humanity at Risk

The National Interest | Boaz Golany

Boaz Golany contends that handing nuclear retaliation to artificial intelligence would strengthen deterrence on paper while removing the human judgement that has averted catastrophe before. North Korea's law of September 8, 2022, which calls for an automatic and immediate strike if its nuclear command system is endangered, prompts him to ask whether Kim Jong-un might delegate that role to AI.

An AI system could monitor warning data, judge whether an attack had occurred and trigger retaliation if no leader could respond. The Soviet Union's Perimeter system, known in the West as the Dead Hand, offered a Cold War precedent. Golany says such autonomy removes any chance of intervention when stakes peak. He cites Stanislav Petrov, who in 1983 judged satellite warnings of incoming US missiles false. Rule-bound software could fail without malfunctioning. Rivals might also speed up their own decisions and lean on automated warning systems. He would keep AI in support roles under meaningful human control.

Comment

Pyongyang's 2022 nuclear forces law turns a declaratory posture into a conditional commitment. Its automatic-strike clause is triggered by danger to the command-and-control system, not by a confirmed attack on North Korean territory. That wording matters because command networks can be degraded by jamming or intrusion well short of nuclear use. A doctrine keyed to the health of the network therefore ties North Korea's nuclear threshold to the reliability of its own sensors and links.

The knock-on effect falls on Pyongyang's adversaries, who would face a reading problem. Any strike on leadership or communications sites could count as the trigger. A conventional raid on a command bunker might then carry unintended nuclear consequences. That uncertainty could push them towards deliberate ambiguity about targeting North Korean command sites, which would blur the clarity deterrence depends on.

Perimeter, the Soviet system known in the West as Dead Hand, rested on a similar link between a damaged command network and retaliation. Stanislav Petrov's 1983 judgement that satellite warnings of US missiles were false shows what that link omits: a person weighing the evidence.

Strategic Question for Discussion

If North Korea's 2022 automatic-strike clause were tied to machine-assessed warning data, which carries more weight, the credibility gained or the control surrendered, and what would tip the balance?

The pattern suggests the credibility gain is real but narrow, because it works only on an adversary convinced the system will behave as designed. Petrov's 1983 judgement points to the opposite risk, that a rule-bound system cannot treat a false warning as one. My assessment is that control surrendered weighs more once rivals begin compressing their own decision time in response.

Share your assessment in the comments below.

To read in details click here