Why Military AI Now Worries Generals More Than Nuclear Weapons: It's Too Fast for Humans to Decide

Nuclear weapons have been with us for 80 years, long enough that we treat them as the ultimate sword hanging over humanity. But at an international defense forum that wrapped up in late May, when experts debated what's most dangerous, the answer had shifted.

Ranked ahead of nuclear weapons: military AI.

A surprising consensus

The forum was the Shangri-La Dialogue, held in Singapore from May 29 to 31 — one of Asia's largest defense and security gatherings, drawing top military officials and strategists from more than 40 countries.

In a session on "strategic stability," the panelists were strikingly aligned: the risks posed by military AI now eclipse those of nuclear weapons.

Lieutenant General Numan Zakaria of the Pakistan Army put it plainly. When AI squeezes reaction time too short, "people don't have enough time to make sense of the situation" — and the result is that "humans take irrational, extreme actions."

Notice what they are not afraid of: AI turning into a sci-fi Skynet. The real worry is far more mundane. It comes down to a single word: speed.

Why "too fast" is such a serious problem

Militaries rely on a classic decision model called the OODA loop: observe, orient, decide, act.

When humans run those four steps, it's slow — but slow leaves room. A missile takes minutes to travel from launch to impact, and in those minutes the officer on duty can double-check, call to confirm, or call the whole thing off. More than one near-miss nuclear crisis was ultimately defused because a human hit the brakes at the last possible moment.

What AI does is compress those four steps into seconds — or less.

The machine observes for you, judges for you, and recommends the next strike. How fast does it get? Fast enough that humans are left only to nod along. Experts call this war at "machine speed" — decisions made so quickly that the human brain can't get a word in.

Once the tempo reaches that point, the danger begins: the machine can miscalculate, mistaking a flock of geese for an incoming squadron. The human, who was supposed to be the firewall at the cliff's edge, fails — simply because there's no time to react. By the time anyone catches up, events may already be unstoppable.

Why military AI is trickier than the bomb

The logic of nuclear weapons is, paradoxically, slow. Their deterrence rests on "no one dares strike first," and that standoff buys time to think and to negotiate.

AI inverts the whole equation. Its entire appeal is speed. Whoever's system reacts a second slower may lose — so every country feels pressure to hand more and more of the decision over to the machine. That race to be faster is exactly where things are most likely to go wrong.

There's a more immediate layer, too: AI is mass-producing information that's hard to tell true from false. Slip a piece of fabricated intelligence or a doctored clip into a tense standoff, add the pressure to decide within seconds, and the odds of a catastrophic misjudgment only climb.

It isn't all bad news

The good news is that the problem is now squarely on the table.

Several nuclear powers — the US, the UK, and France — have in recent years publicly committed to one red line: the decision to launch nuclear weapons must be made by a human, never delegated to AI. International dialogue on "responsible use of AI in the military domain" is ongoing.

Keeping AI away from that single most dangerous button is the minimum consensus everyone can currently reach.

The Shangri-La Dialogue produced no final answer. But it at least stated the problem clearly: what keeps these generals up at night is whether we've left ourselves those few minutes to call it off.