Beyond the AI Nuclear Summit: The Military Reality in AI at War

Picture the Pontifical Gardens at Castel Gandolfo in 2026: clipped hedges, a slope down toward the lake, and inside Borgo Laudato si', a room where Nobel laureates, world leaders, and AI researchers have gathered to talk about artificial intelligence and nuclear war. The Vatican convening the Global Nobel Laureates Assembly on that pairing is not a quiet gesture. It tells you the people closest to the technology worry about the same thing the rest of us half-notice between headlines. Moral urgency and operational understanding are different animals. A cardinal blessing a summit tells you a fear has become respectable. It says nothing about what an autonomous targeting system does when the network drops, or where a commander is still legally required to keep a human finger on the trigger. That second question is harder, and it is much quieter.

The easy assumption is that military artificial intelligence is just consumer AI with a uniform on: the same models, scaled up, pointed at something dangerous. So the debate collapses into a single lever. Either you trust the machines or you fear them. That framing is comfortable and mostly wrong. The systems entering a warship's combat information center train on sparse, contested, deliberately poisoned data, and they break in ways your phone's autocomplete never will. A summit can agree the stakes are high while nobody in the room can say what a sensor-fusion algorithm actually does at three in the morning inside a jammed electromagnetic environment. That distance, between the ethical conversation and the engineering reality, is where headline coverage on artificial intelligence news tends to stop just as things turn concrete.

Sam J. Tangredi and George Galdorisi assembled more than thirty contributors for AI at War, and the roster itself makes the case. Former Defense Department officials sit beside retired flag officers, working scientists, and active-duty junior officers who will one day operate these systems. That mix produces friction instead of a single house line, which is the honest way to treat a subject where nobody yet has settled answers. The book plants its claims in the U.S. Navy and Marine Corps, and the specificity earns its keep. Instead of waving at the future of warfare, it works through what integration means inside particular warfighting functions: how a system travels from a lab demonstration to a fielded capability, what the acquisition process does to it on the way, and where doctrine has to be rewritten before anyone trusts the hardware.

Its steadiest work is drawing the line between what military AI can do and what it still cannot. The technology processes sensor data faster than a human watch team, flags patterns in surveillance feeds, and manages tasks too tedious or too fast for people. It remains poor at the things that matter most under fire: judgment in situations it has never seen, resistance to an adversary feeding it lies, and any account of its own reasoning that a commander could defend afterward. On ethics the book stays practical, and this is where it is both strongest and most likely to frustrate you.

It treats human control as a design and doctrine problem rather than a slogan, asking where accountability attaches once an intelligent system sits in the decision loop. That is the right level of seriousness. The measured, professional register that keeps the analysis credible also files down the moral edges. A chapter written by someone still inside the institution can only push so hard against that institution, and you can feel that ceiling in a few places where the argument stops just short of saying what it clearly wants to say. The payoff is a working vocabulary for the argument happening at Castel Gandolfo and everywhere else.

When a laureate warns about AI speeding a path toward nuclear use, this book lets you ask the follow-up: speeding which function, at what stage of the decision, with what human check still standing. It swaps alarm for judgment, which is less thrilling and far more useful.

If the Castel Gandolfo gathering has your attention, this is a good place to spend it on something concrete. AI at War will not reassure you that the experts have it handled, and it will not hand you a clean verdict to repeat over dinner. It gives you the texture of the actual problem, worked through by people who disagree with each other in useful ways. Sit with it if you want the version of this debate that outlasts the summit's closing statement.