The Pentagon wants AI in its nuclear command, but won't let it push the button

Hollywood made an entire franchise out of what happens when computers get control of America's nukes. But at the Pentagon, commanders are now arguing that keeping AI out of the nuclear enterprise might be just as dangerous.

At the Air & Space Forces Association’s 2025 Warfare Symposium, Air Force and Space Force leadership made a direct pitch: artificial intelligence needs to be woven into the nation’s Nuclear Command, Control, and Communications (NC3) network. Maj. Gen. Ty Neuman of Air Force Global Strike Command put it bluntly, saying that if the military doesn't consider AI, they are going to lose, and he has no interest in losing.

Finding a secure line through the noise

The pitch is less about letting algorithms call the shots and more about keeping the plumbing running when seconds count. In a real-world nuclear scenario, the president has mere minutes to identify a threat, evaluate options, and issue orders to the nuclear triad of bombers, silo-based missiles, and submarines.

Neuman argued that AI is essential for routing those critical messages under fire. If the National Command Authority transmits a strike order, signals might scatter in a hundred different directions across contested airwaves. Some channels may be jammed or compromised. A human operator sitting at a console simply cannot calculate the safest, fastest transmission pathway in real time, but an AI model can.

Space Force Col. Ryan Rose pointed to other unglamorous but vital tasks, including predictive maintenance to prevent unexpected system outages and scanning network traffic for adversary cyber activity. STRATCOM chief Gen. Anthony Cotton echoed similar thoughts in late 2024, noting that refusing to use AI to crunch terabytes of data would be like moving out of modern command bunkers and going back to rotary phones.

Keeping the president on the trigger

Every officer on the panel rushed to emphasize the ultimate guardrail: humans will always be in or on the loop. Machines will never have the authority to launch weapons.

Lt. Gen. Andrew Gebara reiterated that only the president of the United States can order a nuclear strike. Neuman also acknowledged the severe technical risks, noting that an AI system is only as good as the data fed into it. If incoming sensor feeds are spoofed or corrupted, the machine’s output cannot be trusted without a human validating it first.

Why it matters

We think the military’s framing here is tactical and disciplined. Instead of pitching a sci-fi digital battlefield commander, officials are treating AI as a high-speed router and an automated maintenance technician. In an era of intense electronic warfare, asking human operators to manually untangle compromised communications networks during a missile crisis is a recipe for failure, so handing pathfinding over to algorithms looks like a smart move.

However, the real danger lies in moving the attack surface from physical antennas to data pipelines. If the communication routing relies on machine learning models, adversaries will inevitably focus on data poisoning, adversarial noise, and sensor spoofing. An AI that selects the wrong channel because of manipulated training data could effectively sever command authority during the only ten minutes that matter.

This marks a clear line in defense tech: AI is moving rapidly into high-stakes strategic infrastructure, but everyone agrees the red button remains strictly analog and human. Expect defense contractors to pivot hard toward verified, tamper-proof AI architectures for secure communications.