US AI Tools in the Iran War: What CENTCOM Confirmed

AI has gone from experimental military software to operational warfare. U.S. Central Command said it had used the most advanced artificial intelligence to scan battlefield information and speed decision-making by U.S. forces in the 2026 war with Iran. Further disclosures from the Pentagon clarified the picture: the U.S. targeting architecture had incorporated the Maven Smart System (a Palantir platform), computer-vision algorithms, large language models, and automated data-fusion tools. That’s the difference; AI didn’t publicly replace commanders or weapons officers. Instead, it reduced the time between sensing, intelligence analysis, target nomination, weapon selection, and battlefield assessment.

CENTCOM Publicly Confirms AI Use

On 11 March 2026, Adm. Brad Cooper, commander of U.S. Central Command (CENTCOM), delivered an operational update on Operation Epic Fury. As a result, CENTCOM has been open about US personnel deploying advanced AI capabilities in the campaign. Cooper said that artificial intelligence can process huge amounts of information rapidly, but humans still make the decisions.

The difference is of technical significance. Modern militaries generate giant volumes of data, from synthetic aperture radar to electro-optical satellites and infrared sensors, from MQ-9 surveillance aircraft to RC-135 signals intelligence platforms, and from electronic support measures to ground radars and other intelligence sources.

Human analysts cannot manually sift through every frame, emitter report, or intelligence message at operational speed. So AI is basically an information processing multiplier. Algorithms can recognize objects, link tracks together, compare them with past intelligence, and point out anomalous activity. The results are presented to analysts and commanders, not to replace the targeting officer.

Maven Smart System: The Digital Backbone

The most prominent publicly known system was the Pentagon’s Maven Smart System (MSS). Maven began as Project Maven, a 2017 effort to apply machine learning and computer vision to analyze military intelligence. By 2026, it had grown to be much more than just an imagery-recognition program. CSIS says the MSS integrates data from various intelligence and operational sources into a unified battlespace interface. By 2024, CENTCOM was already utilizing 179 different sources of data, and that number was said to be increasing.

The architecture is designed to solve a basic problem of modern warfare: Sensors often generate information faster than analysts can consume it. Computer vision algorithms can automatically identify vehicles, aircraft, missile launchers, or other objects in images from satellites and drones. These detections can then be plotted geographically in the Maven operational interface. They can be compared to other intelligence, and then operators can decide if they want to continue surveillance or military action. This capability changes the traditional find, fix, track, target, engage, and assess (F2T2EA) kill chain.

From Eight Systems to One Interface

One of Maven’s most important capabilities is data fusion. In the past, targeters may have had to manually move information between intelligence databases, imagery terminals, mission-planning applications, and weapons-management systems. “We can take functions that were scattered across, say, eight or nine separate systems and put them in Maven,” Cameron Stanley, the Pentagon’s chief digital and AI officer, said later. The platform could assist personnel from identifying a potential target, through developing a course of action, to matching appropriate friendly assets against that target.

This capability is important from an operational perspective because, often, the speed of decision matters more than raw weapon performance for combat effectiveness. For example, a transporter of ballistic missiles may only remain stationary for short periods. The less time it takes to detect, classify, validate, and assign a weapon, the more likely the target will still be in the engagement window. The big thing AI is doing to warfare is reducing latency.

Claude and Generative AI

Maven also has generative AI capabilities. Palantir’s Artificial Intelligence Platform adds a layer for large language models to interact with operational databases. “Claude from Anthropic was important in particular because they had already embedded it in classified U.S. government networks with Palantir. CSIS states that Claude supported the Iran campaign and provided Maven’s large language model capabilities. We must distinguish traditional machine learning from an LLM.

Computer vision systems can detect a launcher in an image. Language models can help analysts search reports, summarize intelligence, compare multiple detections, retrieve supporting information, and explore possible courses of action using natural-language commands. This process creates the following layers in the AI architecture:

sensors → computer vision → data fusion → LLM-assisted analysis → human review → command decision → weapon employment.

Public evidence does not establish that Claude autonomously ordered strikes. CENTCOM specifically emphasized that human responsibility for engagement decisions remained in place.

Case Study: 1,000 Targets in 24 Hours

The scale of operations showed why the Pentagon was paying attention to AI. In the first 24 hours of the Iran campaign, U.S. forces used the technology to strike more than 1,000 targets, CSIS said. It characterized the figure as a rough tenfold increase in targeting throughput compared to what was possible before MSS-era capabilities. As of March 16, CENTCOM said it had conducted more than 7,000 strikes and 6,500 combat flights as part of Operation Epic Fury.

The force included B-1, B-2, and B-52 bombers; F-15, F-16, F/A-18, F-22, and F-35 fighters; MQ-9 drones; U-2 and RC-135 surveillance aircraft; and electronic warfare, sea, and land-based systems. The density of operations creates a massive command and control problem.” Thousands of sorties, sensor feeds, target coordinates, airspace restrictions, weapons inventories, and threat emitters must remain in sync. Therefore, the strategic value of AI ranges from target recognition to battlespace management and decision synchronization.

The Extraordinary Scale of AI Processing

The Pentagon figures showed how dramatically the use of AI had expanded in the postwar period. During the 38-day campaign, Maven supported operations involving about 13,000 targets, the Pentagon said. Maven use in classified applications increased 89%, and unclassified use increased 38% over the previous month. The most striking statistic was about generative AI.

Peak consumption hit nearly 20 billion AI tokens per day, the Pentagon said—a 4,425% jump in peak daily usage. Language models read prompts and output answers using tokens as their base computational units. Billions of tokens means continuous machine processing with many operational users, not the odd chatbot-like query. This meant that generative AI was no longer an experimental application but was integrated into the military staff workflow.

What Remains Classified

Most of the architecture is still secret. The CENTCOM fact sheets themselves acknowledged other “special capabilities” without specifying them. The model configurations, the weighting of intelligence sources, the classified sensor interfaces, the confidence thresholds, the electronic-intelligence algorithms, and the detailed rules for automated target recommendations are not public. Therefore, we should approach new autonomous target systems with caution. The most important publicly known advancement is already substantial: the US demonstrated that AI could integrate ISR data, prioritize information, support targeting personnel, and cut military decision cycles in a major interstate war.

Open-source intelligence (OSINT) analysis has identified several specific platforms depicted as nodes within the broader MSS architecture. Such nodes’ operational involvement had not previously been publicly identified. Official assets listed include LUCAS drones and EC-130H Compass Call electronic warfare aircraft. The intelligence, surveillance, reconnaissance, and communications fleet included the RC-135, U-2, P-8, E-2D Hawkeye, and specialized airborne communications-relay aircraft. Separately reporting says E-11 BACN (Battlefield Airborne Communications Node) aircraft were operating in the area, providing high-altitude communications bridging and data-relay support across dispersed U.S. and allied forces.

Human Control Remains the Critical Layer

However, there are risks to AI-assisted targeting. Machine-learning systems can evaluate false, old, or incomplete intelligence very rapidly. More processing speed does not necessarily mean more intelligence accuracy. That is a simple paradox. Artificial intelligence may provide commanders with additional information but less time to study it. The critical question is not just whether humans remain “in the loop” but also whether they have the time, context, and authority to push back on machine-generated recommendations.

What matters is whether commanders have the time, context, and authority to push back on machine-generated recommendations. Operation Epic Fury was more than just another deployment of sophisticated weapons. It showcased a new type of warfare in which the key advantage may increasingly lie in the ability to process large amounts of sensor data into actionable military options faster than an adversary can react. The aircraft, the missiles, and the drones were kinetic instruments. They were tied more and more together by the connective tissue that was AI.

References

1. U.S. Central Command Public Affairs, “CENTCOM Commander Provides Update on Operation Epic Fury,” March 11, 2026.
2. U.S. Central Command, Operation Epic Fury Fact Sheet, March 16, 2026.
3. Matt Mande and Gregory C. Allen, CSIS, “What Is Maven Smart System, and What Does It Do?”, June 2, 2026.
4. Sydney J. Freedberg Jr., Breaking Defence, “‘Insatiable appetite’ for AI: Maven usage surged for strikes on Iran,” May 12, 2026

About the Author

Farhan J. Satti (Chief Editor)

Administrator

Mr. Farhan J. Satti is a defense analyst and HR leader with 15+ years covering weapons, arms control, and operations. UN Disarmament Affairs-trained (WMD missiles, lethal autonomous weapons) and firearms-safety trained (Washington State), he explains hard problems with clarity and impact.

Visit Website View All Posts