Geostrategic Media Political Commentary, Analysis, Security, Defense
Hadi Elis
The United States entered its war with Iran possessing something few militaries in history have had at comparable scale: extraordinary technological superiority, an enormous intelligence architecture and increasingly sophisticated artificial-intelligence systems capable of processing information at speeds no human organization could match. Artificial intelligence has become deeply embedded in modern American military operations, from intelligence fusion and surveillance to target identification, prioritization, battle management and operational planning. During the war in Iran, U.S. forces reportedly relied on the Maven Smart System to integrate information from satellites, drones, radar, communications and other sources and to help identify and prioritize targets. The speed was remarkable. But speed is not the same thing as understanding. And this may be the most important lesson emerging from the war: a military can possess more data, process it faster and generate more targeting options than its opponent, yet still struggle to understand what the opponent is actually trying to achieve.
This distinction matters because war is not ultimately a data-processing competition. It is a contest between political systems, institutions, societies, perceptions, fears and calculations. An algorithm can identify a missile facility, a command centre, a radar installation or a senior military figure. It can compare thousands of pieces of information and rapidly produce a list of potential targets. It can even assist commanders in selecting weapons, attack routes and courses of action. But the most difficult questions in war are often not geographical or technical. They are political. What happens after a leader is removed? What does the adversary consider an existential threat? Which institutions survive the destruction of a particular command structure? Which actors gain power when others disappear? What does the population believe is happening? Which military losses will produce capitulation, and which will instead produce greater resistance? These are questions about human behaviour and political legitimacy, not merely about data.
Iran presents precisely this problem. Its political system cannot be understood simply through the formal institutions visible on an organizational chart. There is an elected president, parliament and other state institutions, but there are also powerful unelected institutions, security structures, religious authorities and networks of influence that shape the country’s strategic decisions. The distinction between the government that appears formally responsible for policy and the deeper architecture of power is therefore essential to understanding Iranian behaviour. An AI system can be fed enormous quantities of information about Iran, but the quantity of information available to a machine does not guarantee that the machine possesses the missing information required to understand the political logic behind that information. In intelligence analysis, what is absent can be as important as what is present.
This is where the idea of AI as an independent military strategist becomes problematic. Artificial intelligence can calculate probabilities, identify patterns and generate recommendations, but it does not possess political intuition in the human sense. It does not experience fear, humiliation, ideology, historical memory or national trauma. It does not understand why a political leadership might prefer prolonged resistance over an apparently rational concession. Nor does it inherently understand the symbolic meaning that human beings attach to a leader, a religious institution, a military organization or national territory. These limitations become particularly consequential in a conflict involving a political system as complex and ideologically structured as Iran.
The American military’s achievements with AI should nevertheless not be dismissed. The Pentagon has spent years building the infrastructure required to make military AI operationally useful. The Brennan Center estimates that the Defense Department has allocated at least $75 billion to AI-related programs since 2016, although the actual figure could be considerably higher because classified programs and programs whose AI components are difficult to identify are not fully captured. The Maven system itself evolved from an initiative launched in 2017 to use machine learning and computer vision to process battlefield imagery. By 2026, it had become considerably more ambitious, integrating multiple sources of information and generating operational recommendations.
The significance of this transformation should not be underestimated. Traditional military intelligence required analysts and commanders to move between different information systems, interpret reports and construct a picture of the battlefield. AI can compress much of that process into seconds. During the Iran campaign, U.S. officials described systems capable of processing vast amounts of information and helping commanders identify targets and potential courses of action. The result is a dramatic acceleration of the targeting cycle. A system can identify a target, compare it with other targets, recommend a weapon and produce a battle plan much faster than a traditional human analytical structure. In a high-intensity conflict, that speed can have genuine operational value.
But acceleration creates a paradox. The faster a machine produces an answer, the less time humans may have to question it. A commander presented with a sophisticated algorithmic recommendation, backed by enormous quantities of apparently authoritative data, may become psychologically inclined to accept the recommendation rather than challenge it. This phenomenon is commonly described as automation bias. The human remains formally responsible for the decision, but the practical centre of decision-making can gradually shift toward the machine. The human being becomes the person who approves the algorithm rather than the person who independently evaluates it. Analysts have warned precisely about this danger in the context of AI-assisted targeting in Iran.
The tragedy surrounding the strike on the Shajareh Tayyebeh primary school in Minab illustrates why this distinction matters, although it would be premature to attribute that incident simply to AI. The school was struck on February 28, with more than 170 people reportedly killed, most of them children. Subsequent reporting cited intelligence and mapping failures in which the facility’s earlier military association apparently contributed to its presence on a target list despite its later civilian use. Whether and to what extent AI contributed to the specific targeting decision remains contested; reporting cited by the Arms Control Association indicates that U.S. officials considered human intelligence and mapping error central to the incident. The lesson, therefore, is not that a machine independently decided to bomb a school. The more important lesson is that when machine-generated recommendations interact with outdated information, compressed decision times and human assumptions, technological sophistication does not eliminate the possibility of catastrophic error.
This is the fundamental weakness of an AI-first conception of warfare. Artificial intelligence is extraordinarily good at processing the world represented in its data. It is much less reliable when the central problem is that the representation itself is incomplete, outdated or politically misleading. If an intelligence database contains an obsolete map, the algorithm can process the obsolete map with extraordinary efficiency. If analysts misunderstand the structure of an adversary’s political leadership, AI can accelerate that misunderstanding. If the underlying assumptions are wrong, greater computational power may simply produce a more sophisticated version of the wrong answer.
Iran also demonstrates why decapitation strategies are not necessarily equivalent to political transformation. Removing senior officials or destroying command centres can produce significant military effects, but the destruction of individuals does not automatically destroy the institutions, networks and ideologies that produced them. Political organizations can be more resilient than their visible leadership. Revolutionary Guards, religious networks, bureaucratic institutions and security organizations may continue functioning even after substantial losses at the top. A targeting system may therefore correctly identify a particular individual as important while simultaneously misunderstanding the individual’s actual position within the broader political system.
This is not unique to Iran. History repeatedly demonstrates that states and movements rarely behave according to the neat logic of organizational charts. Saddam Hussein’s removal did not eliminate the political and social forces that shaped Iraq. The killing of Osama bin Laden did not eliminate jihadist ideology. The defeat of one military formation does not necessarily remove the political conditions that created it. Modern military planners understand this intellectually, but the enormous power of AI-generated targeting can create a temptation to reduce political complexity to identifiable nodes. Once a problem is translated into a network diagram, there is a natural temptation to believe that destroying enough nodes will cause the network to collapse.
The danger becomes greater when AI moves from decision support toward autonomous or semi-autonomous action. There is an important difference between a system that helps an analyst discover a possible target and a system that effectively determines which target should be attacked. The first can augment human intelligence; the second risks transferring strategic authority to a technology that does not understand the political consequences of its recommendations. The Pentagon itself has been confronting this question. The dispute between the Defense Department and Anthropic over restrictions on the use of Claude in fully autonomous weapons highlighted the unresolved tension between technological capability and human control. The Brennan Center has also warned that military AI can displace human expertise and judgment precisely when decisions involve the lives of soldiers and civilians.
The problem is not simply ethical. It is strategic. A machine has no political responsibility. It cannot be held accountable for misreading an adversary’s intentions. It cannot understand why an apparently minor symbolic event might transform public opinion. It cannot experience the political consequences of civilian casualties. It cannot independently determine whether destroying a particular target will shorten a war or instead deepen the enemy’s determination to continue it. These are precisely the questions that distinguish tactics from strategy.
The Iran war has also demonstrated that AI is transforming not only the physical battlefield but the information battlefield. Generative AI has made it possible to produce convincing images, videos and narratives at extraordinary speed. Research from Brookings found a significant increase in AI-related misinformation and fabricated material surrounding the conflict, including recycled footage, fabricated attacks and synthetic imagery. More than 5,000 Community Notes on X referenced AI-generated content during the period examined by the study, although that represented only a fraction of the broader information ecosystem and did not establish that every flagged item was actually AI-generated.
This creates another paradox. The same technology designed to help a military see the battlefield more clearly can contribute to an information environment in which seeing reality becomes more difficult. Artificial intelligence can process satellite imagery, intercepts and sensor data, while simultaneously generating convincing false images that confuse the public, journalists and even analysts. The battlefield therefore becomes not only a contest over territory and firepower but a contest over epistemology: who can establish what is real before the next narrative replaces it?
That may ultimately be the most important lesson of the American experience with military AI in Iran. The problem is not that AI has no place in war. It clearly does. Nor is the evidence sufficient to conclude that AI alone caused American strategic difficulties or that the technology itself “failed” in the war. The evidence points to something more complicated. AI has demonstrated considerable value as an instrument for processing information, accelerating intelligence analysis and supporting military decision-making. At the same time, the war illustrates the danger of confusing operational efficiency with strategic comprehension.
A successful military strategy requires more than knowing where the enemy is. It requires understanding why the enemy is there, what the enemy believes it can survive, what it fears losing, what it is willing to sacrifice and what political outcome it considers preferable to surrender. Those questions cannot be answered simply by adding more computing power. They require historians, diplomats, intelligence officers, regional specialists, political scientists, military commanders and people who understand the adversary’s language, institutions and culture. They require disagreement inside the decision-making process rather than the illusion that an algorithm has produced the single correct answer.
The United States therefore faces a choice that is larger than the question of which AI model it uses. It can treat artificial intelligence as an extraordinarily powerful assistant to human strategy, or it can gradually allow technological systems to become the hidden authors of strategic decisions. The first approach recognizes the strengths of machines while preserving the weaknesses—and responsibilities—of human judgment. The second risks creating a dangerous illusion: that because a decision has been generated from billions of data points, it must therefore be strategically intelligent.
War is not a spreadsheet. An adversary is not a dataset. A political system is not merely a network of targets. And destroying the visible architecture of power does not necessarily mean understanding the power that lies beneath it.
The United States may ultimately conclude that the central lesson of the Iran war is not that artificial intelligence is incapable of fighting wars. It is that AI is exceptionally capable of fighting the part of war that can be converted into data. Strategy begins where the data stops.
Autho’s Bio:
Hadi Elis is a Kurdish Canadian writer based in Ontario.
