During my lifetime, national security meant managing threats we could see and count. We tracked enemy nuclear stockpiles, built command-and-control systems, and used diplomacy, military readiness, and deterrence to prevent a crisis. Nuclear safeguards didn’t erase the danger, but leaders understood that controlling a catastrophic technology could not be an afterthought.
AI is testing whether we still remember that lesson. It is changing how we gather intelligence, react to cyber threats, and plan military operations. AI is already part of our national defense. The question is who will control it.
I spent much of my military career in U.S. Army light infantry and Special Forces units. The military depends on a clear chain of command. Someone has to be answerable for the final decision, especially when an algorithm can act faster than any human operator.
When it comes to using a nuclear weapon, humans remain accountable for the decision. Those protections exist because a mistake involving nuclear weapons cannot be corrected afterward.
AI raises a similar question: What happens if we lose control?
The question has reached Capitol Hill. The AI Guardrails Act would put legal limits on the Pentagon’s use of AI, including decisions involving nuclear weapons and lethal force. Sen. Angus King (ME), a member of the Armed Services Committee, has argued that America needs to move into an artificial intelligence future “carefully, warily, and thoughtfully.”
That warning proved prophetic on Friday. CNN reported that an AI-generated intelligence report falsely flagged a Chinese cargo ship in the Middle East as carrying nuclear weapon components. Armed service members were readying to board, and military aircraft were in the air when intelligence officials took a closer look and discovered the AI-assisted report was wrong. One source told CNN the incident “almost started a war.”
That is the kind of mistake that concerns me. In this case, humans caught it. What happens when we don’t? We didn’t wait until missiles were in the air to build missile defenses. We prepare because the consequences of getting it wrong are too great.
If we put AI into early-warning or military decision systems, leaders may have less time to figure out whether the information in front of them is right.
Nuclear weapons and artificial intelligence are different in another important way. Governments built the atomic bomb and tightly controlled the nuclear material needed to make it. Much of today’s most advanced AI is being built by private companies. We want American engineers to lead, but corporate leadership shouldn’t be confused with democratic accountability.
Silicon Valley has long celebrated moving fast and breaking things. It is different when lives are on the line. Neither tech CEOs nor software engineers should dictate military command structures or the rules for autonomous lethal force.
Critics argue that statutory limits will cause the United States to fall behind adversaries such as China, pointing to existing Pentagon policies as sufficient. Internal military guidelines alone cannot answer how much authority we are willing to hand over to machines. Some decisions should never be outsourced to code. Congress must put those limits into law. AI should not make decisions involving nuclear weapons or other strategic systems without human control.
The U.S. needs to talk with other major powers about where those boundaries should be. We don’t need to begin with a sweeping global treaty. Start with the decisions that are simply too dangerous to leave to a machine: launching a nuclear weapon, controlling strategic defense systems, or automatically responding to an early warning.
History gives us an example. In September 1983, a Soviet early-warning system falsely reported incoming American nuclear missiles. Soviet officer Stanislav Petrov concluded that the warning was a system error rather than an attack. His judgment mattered. Had an automated system acted on that warning without intervention, the consequences could have been catastrophic.
More than four decades separate Petrov from what CNN reported this past week. Technology has changed. The concern remains. Machines can process information and identify patterns we miss, but they can also be wrong. What worries me isn’t that a machine someday becomes smarter than us. It’s about whether we start trusting machines to make decisions simply because they are fast.
AI DIDN’T ALMOST TRIGGER A WAR WITH CHINA. WASHINGTON’S BLIND FAITH IN IT DID
During my military service, the question of who had the authority to send Americans to war was never abstract. The responsibility of deciding when Americans should fight belongs to our elected leaders.
AI changes that question, but it doesn’t change who should be accountable. As AI speeds up decisions in military actions, we need to decide where the machine stops and human ownership begins. When the stakes are this high, we can’t wait for a crisis to draw that line.
Scott W. Patton is a combat veteran and Gold Star family member whose work has appeared in USA Today, the Chicago Tribune, HuffPost, The Plain Dealer, the Washington Examiner, and other national publications.