The next great military confrontation may begin long before the first missile launches, the first aircraft crosses a border, or the first soldier receives orders to advance.

It may begin inside competing computer networks, where artificial intelligence systems help evaluate surveillance imagery, interpret electronic signals, predict possible enemy intentions, and recommend responses at speeds no human commander can match.

For centuries, warfare has been shaped by a fundamental limitation: the speed at which human beings can acquire information, understand it, and make decisions. Artificial intelligence is changing that equation. But faster decisions do not necessarily mean better decisions.

From the Commander’s Assistant to the Commander’s Dilemma

Information reaches a modern command center from satellites, aircraft, radar installations, ships, intelligence agencies, and troops in the field. Some reports reinforce one another; others conflict. Some are outdated before they reach the commander. The challenge is no longer simply obtaining information. It is determining which information matters.

AI offers substantial opportunities. It can help analysts identify patterns, compare conflicting reports, flag unusual activity, and present commanders with a more coherent picture. It may help identify hazards along flight corridors, highlight potential friendly-force positions, and draw attention to uncertainty in a proposed operation.

But a recommendation that arrives in seconds can require much longer to verify. The danger begins when the assistant becomes so fast, complex, and apparently reliable that commanders lack the time or evidence to challenge its conclusions.

When Both Sides Have Artificial Intelligence

Imagine two opposing military forces, each using sophisticated AI decision-support systems. One detects activity it interprets as preparation for an attack and recommends increased readiness. The other detects that increased readiness and interprets it as hostile intent. It recommends a countermeasure. The first detects the countermeasure and raises its assessment of the threat.

Neither leadership intended to start a war. Yet each system has helped create the conditions for escalation. This is a hypothetical scenario, not a claim that today’s military AI independently commands such actions. It illustrates a real strategic concern: faster interactions can compress the time available for human judgment.

The Human Brain Is Not a Ten-Percent Machine

The popular claim that people use only ten percent of their brains is a myth. Human beings use networks throughout the brain. The real limitation is different: attention is finite, working memory is constrained, fatigue affects performance, and a decision-maker cannot independently examine thousands of rapidly changing signals at once.

AI can expand what a commander can observe. It cannot automatically resolve questions of political intent, legal authority, moral responsibility, or whether an apparent tactical advantage is worth the strategic consequences.

Deception at Machine Speed

Intelligence has always involved uncertainty and deception. Camouflage, false communications, decoys, misleading movements, and electronic interference can distort an adversary’s understanding. AI adds another layer: systems may process manipulated data at scale and produce persuasive conclusions from unreliable inputs.

A model can be technically functioning as designed and still be wrong because the information supplied to it is incomplete, deceptive, or out of date. The critical question becomes not merely, “What does the system recommend?” but “What evidence supports that recommendation, and what evidence would disprove it?”

Intelligence, Counterintelligence, and the False Premise

The fundamentals of intelligence and counterintelligence have not disappeared with the arrival of AI. Intelligence seeks to establish an adversary’s capabilities, activities, and intentions. Counterintelligence seeks to detect, frustrate, and understand hostile efforts to collect information, penetrate systems, and deceive decision-makers. Both disciplines depend on skepticism, source evaluation, corroboration, and a willingness to reconsider an initial judgment.

AI is not infallible. It can overlook context, misclassify information, and express unwarranted confidence. A technically sound model can produce a logically consistent answer to a false premise. Worse, an adversary who understands which signals a system treats as significant may attempt to shape those signals. That possibility does not prove an attack has occurred; it demands that analysts test for one.

Intelligence professionals distinguish the reliability of a source from the credibility of a particular report and from the confidence warranted by the overall assessment. A machine-generated conclusion should face the same scrutiny. Question the source. Challenge the assumption. Test the alternative.

The Difference Between Targeting and Judgment

AI-assisted analysis may help identify objects, estimate risks, or compare courses of action. But identifying a possible target is not the same as establishing that an attack would be lawful, necessary, or strategically wise. Those assessments require reliable information, applicable legal standards, command authority, and human accountability.

Automation bias—the tendency to place excessive trust in a system’s output—can make the danger worse. A commander who is expected to approve an opaque recommendation almost instantly may retain formal authority without exercising meaningful control.

The Case for AI—and the Case for Restraint

There is a strong argument for using AI in military operations. Better analysis could help prevent misidentification, reduce avoidable losses, detect threats earlier, and provide commanders with more options. Refusing useful technology can carry risks of its own.

There is also a strong argument for limits. Systems can fail in unfamiliar circumstances; adversaries can deliberately mislead them; and rapid, interconnected responses can amplify small errors. Safeguards must be tested under realistic conditions, including degraded communications, conflicting reports, and adversarial deception.

The objective should not be to make human judgment ceremonial. It should be to make human judgment better informed.

The Intelligence Assessment

What We Know

Military organizations are developing and employing AI-enabled tools for analysis and decision support. Human attention and decision time are limited. Intelligence can be incomplete or deceptive. Faster operational tempo can create pressure to act before uncertainty is resolved.

What We Assume

Future adversaries will increasingly use AI-enabled systems to interpret one another’s behavior. Those systems will not always share the same data, objectives, or assumptions. Human leaders may be tempted to rely on recommendations that are difficult to audit in real time.

If / Then Scenarios

• If AI improves the quality and timeliness of intelligence while preserving independent review, then commanders may make safer and more effective decisions.
• If competing systems interpret defensive measures as offensive preparations, then they may intensify a crisis neither side intended.
• If an adversary corrupts the information entering a system, then a plausible recommendation may rest on a false premise.
• If commanders cannot explain or challenge a recommendation before authorizing force, then nominal human approval may provide little real protection.
• If an adversary learns how an AI system weighs particular indicators, then that adversary may attempt to influence its assessments without directly engaging military forces.

Alternative Explanation

AI need not accelerate escalation. Systems that reveal uncertainty, cross-check sources, and warn against premature action could give leaders more time to avoid a conflict. The outcome depends on design, doctrine, training, oversight, and political choices—not on computing speed alone.

What Could Change Our Assessment?

Evidence that military AI systems consistently identify deception, communicate uncertainty clearly, and permit effective human intervention under time pressure would strengthen the case that AI can reduce escalation risks. Documented failures under adversarial conditions, or decision processes in which human review becomes merely nominal, would strengthen the case for greater caution.

Our Assessment

Moderate confidence: AI will increasingly influence the speed and presentation of military decisions. Low confidence: any specific prediction that AI-to-AI interaction will independently trigger a future war. The principal risk is not inevitable machine autonomy; it is the human decision to trust fast, opaque recommendations without adequate scrutiny.

Final Perspective

Winning a battle is not the same as making the right decision to fight it. Artificial intelligence may help commanders see farther, understand more, and respond faster than ever before. But speed is a tool, not a strategy—and an algorithm cannot assume responsibility for the human consequences of war.

The next war may be won before the first shot is fired. The greater danger may come when the human commander can no longer explain why that first shot was necessary.