A record of the 2026 Iran War and the kill chain ruled by artificial intelligence

Chapter 1. Claude's War, the Day Silicon Valley Entered the Kill Chain

At dawn on February 28, 2026, more than 200 fighter jets pushed into the skies over Tehran. Somewhere thousands of kilometers away from the battlefield, a large language model was reading satellite imagery, signals intelligence, and surveillance feeds at the same time. It was Claude, the AI created by Anthropic. Integrated into Palantir's Maven Smart System, Claude was reported to have generated about 1,000 priority targets during the first 24 hours.

According to multiple outlets, including The Washington Post, Reuters, and Yonhap News, Claude was used for intelligence assessment, target identification, and combat scenario simulation. An AI that had helped people with their homework until yesterday was now calculating whom to bomb first.

Palantir Technologies is a company that has long built the digital nervous system of U.S. defense. In July 2025, it signed a 10-year, $10 billion contract with the U.S. Army. Maven-related contracts expanded to $1.275 billion, and more than 20,000 personnel across 35 units were using the system routinely. Palantir's Gotham platform links scattered data such as satellite imagery, communications intercepts, and field reports through ontology technology to find hidden relationships and patterns. Titan is a mobile ground station that sends real-time targeting information to the front line. The Artificial Intelligence Platform, AIP, connected large language models directly to military operations, allowing commanders to ask about situations verbally and receive answers.

Claude was placed on top of this. By synthesizing drone video, satellite data, and communications intercepts in real time, it finished within hours work that in the past would have required thousands of analysts for days. Target selection time fell from hours to minutes.

War looks like a vast noise, but in reality it is a technology of classification. Sorting which signal is a threat. That distinction sets the speed of combat. What Palantir created was a map of data. What the language model changed was the way that map is read. A language model does not bomb images directly. Instead, it creates sentences inside a flood of data. "What risk does this target pose?" "Which launcher still retains counterattack capability now?" "Which strike will collapse the air defense network first?" The commander does not spread out dozens of reports and assemble a puzzle. Ask a question, and the model returns summaries, classifications, priorities, and even operational scenarios. At that moment, the language model is no longer a secretary. It becomes a staff officer.

The problem is that this staff officer does not know fear. Humans hesitate before ambiguity. A model turns even ambiguity into polished sentences. When that fluency looks like trust, the danger begins.

This war is not a story of "AI fired a missile." AI changed the order of the screens commanders see. That is where its character is decided.

Targets are organized faster, and priorities are updated in real time. Human commanders are pushed toward approving from among options compressed by the system rather than judging directly. When a misjudgment occurs, responsibility blurs between human and machine. The persuasiveness of the model's sentences hides the incompleteness of the facts. The old fact remains that war moves through language. Only the writer of that language has changed.

But on that very day, February 28, when the bombing began, a historic collision occurred. Anthropic argued that Claude should not be used for mass surveillance of U.S. citizens or for fully autonomous weapons that exclude human decision-making.

The Pentagon demanded access for "all lawful purposes." President Trump ordered all federal agencies to stop using Anthropic technology. But when that order came down, Claude had already been used for intelligence analysis and target selection in the first hours of the Iran airstrikes.

Anthropic CEO Dario Amodei said in a statement, "I deeply believe it is existentially important to use AI to defend America and democracy, but today's AI systems are not reliable enough to operate fully autonomous weapons." The U.S. Department of Defense designated Anthropic a "supply chain threat." OpenAI's Sam Altman filled the void by signing a separate contract with the Pentagon within hours.

The reaction was immediate. More than 900 technology workers at Google, OpenAI, and other companies signed a solidarity petition. The public was more direct. Within 48 hours of OpenAI's defense contract announcement, 1.5 million paid subscribers left. The deletion rate of the ChatGPT app jumped 295%. On the other side, Claude took the overall number-one spot in Apple's App Store for the first time in its history. Performance alone is not enough. Trust became a variable that determines a company's survival.

Behind this drama lies a colder fact. Whether Claude or OpenAI's models, the deployment of large language models for target generation and intelligence fusion was wholly unprecedented in military history. According to an analysis by The Jerusalem Post, without AI systems, the United States and Israel "would not have achieved more than 5-10% of the target list." This moment, when a technology supplier's conscience collided head-on with the state's capacity to wage war, became the archetype of a structural tension that every AI war will face again.

Chapter 2. Chain Reaction, from the Hamas Attack to the Assassination of Khamenei

To understand today's wars, we have to return to October 7, 2023. Hamas's attack on Israel was the trigger for a geopolitical chain reaction. Iran's Axis of Resistance, Hamas, Hezbollah, the Houthis, and Iraqi militias, pressed Israel at the same time. In April 2024, Iran directly struck the Israeli mainland for the first time in history. It was Operation True Promise. 170 drones, 30 cruise missiles, and 120 ballistic missiles flew at once. It was the largest drone attack in history.

The turning point was June 2025. The IAEA reported that Iran possessed highly enriched uranium sufficient for nine nuclear weapons. Immediately afterward, Israel launched Operation Rising Lion. More than 200 fighter jets dropped more than 330 munitions on about 100 targets across Iran, including Natanz, Isfahan, and Fordow. Precision weapons launched from drone bases the Mossad had secretly built near Tehran months earlier disabled Iran's air defenses.

In this operation, AI was already functioning as a core link in the kill chain. An Israeli intelligence officer interviewed directly by the AP testified that AI had been used since October 2024 to sift through vast surveillance data. It analyzed satellite data, identified targets, and optimized attack routes.

U.S.-made AI systems were also used in parallel to improve accuracy and minimize collateral damage. Targets were classified by category, including leadership, military, civilian, and infrastructure. An intelligence officer was assigned to draw up lists of Iranian generals' workplaces and movements during leisure hours. CSIS analyzed the operation and concluded: "Deep integration of special operations forces, autonomous drones, and AI-enabled intelligence, surveillance, and reconnaissance, ISR, is now the baseline for theater entry." Swarms of small explosive drones first saturated Iranian air-defense radars and communications nodes. Then 200 fighter jets followed with precision strikes.

The result of the first airstrike was harsh. About 30 generals, including IRGC commander-in-chief Hossein Salami, were killed. At least nine nuclear scientists died. The Jerusalem Post assessed that the IDF had used AI and big data in earlier operations as well, but the scale of AI integration in Rising Lion was "unprecedented." During the 12-day war, Iran retaliated with more than 550 ballistic missiles and more than 1,000 Shahed suicide drones. Israel's multilayer air defenses intercepted about 86%, but 29 Israelis were killed. Iranian deaths exceeded 600.

The United States intervened. It was Operation Midnight Hammer. Seven B-2 stealth bombers dropped 12 Massive Ordnance Penetrators on the Fordow underground facility. A ceasefire was declared on June 24. AI and cyberwarfare were decisive in this operation as well.

According to an exclusive report by Recorded Future News, U.S. Cyber Command, CYBERCOM, digitally struck "upstream" nodes of military networks connected to Iranian nuclear facilities. The goal was to prevent Iran from launching surface-to-air missiles. It was the most sophisticated anti-Iran operation in Cyber Command's 16-year history.

Chairman of the Joint Chiefs of Staff General Caine publicly stated that Cyber Command supported the "attack package." Large-scale electronic jamming took place across the electromagnetic spectrum. The result is captured in one line from Caine: "Iranian fighters did not fly, and it appears Iran's surface-to-air missile systems did not see us." While 125 aircraft struck all three nuclear facilities within 30 minutes, Iran was blind.

Meanwhile, AI-based psychological warfare unfolded simultaneously in cyberspace. Researchers at Clemson University and Canada's Citizen Lab uncovered a network of fake X accounts named "PRISONBREAK." An AI-powered manipulation campaign was inciting rebellion among the Iranian public. This campaign, which began in January 2025, peaked during the June war. But the setback to Iran's nuclear program was assessed at only several months to at most two years.

In December 2025, the largest anti-government protests in Iran since the 1979 revolution spread to more than 100 cities. The Revolutionary Guard massacred protesters. Trump warned of military intervention. The largest buildup of Middle East military power since the 2003 Iraq invasion began. Two aircraft carriers, F-15E fighters, and 14 aerial refueling aircraft deployed to Israel's Ben Gurion Airport. It was the first time in history that U.S. offensive weapons were deployed in Israel.

Then February 28, 2026 arrived. Joint U.S.-Israeli airstrikes began. Israel, based on CIA intelligence, independently struck Khamenei's private residence. The supreme leader, his daughter, son-in-law, grandson, and daughter-in-law were killed. About 40 senior Iranian officials died in the initial airstrikes. Iran retaliated with Operation True Promise IV. It launched missiles and drones across nine countries. It blockaded the Strait of Hormuz. A missile hit a synagogue in Beit Shemesh, killing nine people. An airstrike on a girls' school in Hormozgan Province killed 165.

The Jerusalem Post called this war "the first full-scale artificial intelligence war in history." In Rising Lion and Midnight Hammer in June 2025, AI was still an "assistant" in the kill chain. But in target identification, cyberwarfare, autonomous drone operation, and psychological warfare, AI was already operating as a core element across every domain. Operation Epic Fury in 2026 was the result of those seeds blooming into full-scale war.

Chapter 3. Eleven Seconds of Electronic Warfare, the Cyberattack That Blinded Iran

When the operation began on February 28, 2026, Iran's internet connectivity plunged to 1-4% of normal levels. In this operation, described as "the largest cyberattack in history," U.S. Cyber Command, CYBERCOM, was deployed with Space Command as a "first strike." In the words of Chairman of the Joint Chiefs General Caine, it "disrupted, degraded, and blinded Iran's vision, communications, and ability to respond."

The numbers show the speed of this operation. It took AI 11 seconds to identify and classify more than 50,000 signals. Adaptive jamming was deployed. Iran's command system was paralyzed for 47 minutes.

Already in Operation Midnight Hammer in June 2025, CYBERCOM had experience digitally disabling Iran's air-defense systems to secure the path for U.S. bombers to strike nuclear facilities. NSA-supported operators struck "upstream" nodes in military networks and prevented Iran from firing surface-to-air missiles.

Iran also counterattacked. Iranian hacker groups used Google Gemini to collect information on Israeli defense systems, satellite infrastructure, and drone technology. Influence operations using OpenAI models, STORM-2035, were detected. After the June 2025 war, Iran-linked cyberattacks surged by 700%. As of early March 2026, about 60 hacktivist groups were active.

The most unexpected attack was not cyber but physical.

On March 1, 2026, Iran used drones and missiles to strike AWS data centers in the UAE and Bahrain. Two availability zones in the UAE region were paralyzed by fire and power loss. Error rates surged for core cloud services such as EC2, S3, and RDS. Physically destroying the other side's servers to degrade AI capability itself. It was the moment a new logic of war emerged. The AI-based economic strategies pursued by Gulf states faced a new risk: a single bomb on a data center.

Both sides carried out cognitive warfare using deepfakes. The Iranian side created and spread AI-generated scenes of Tel Aviv's virtual destruction. Israel's Unit 8200 developed synthetic media operations, generative AI voice cloning and GAN video. Cases spread in which AI-altered satellite images exaggerated or misled about actual damage. Social platform X had to introduce a policy "restricting monetization when AI-generated war videos are posted without labeling." The boundary between war and falsehood was melting under AI.

Chapter 4. Lavender, Gospel, and Where's Daddy?, Israel's Algorithmic Killing Factory

The most controversial and most efficient technologies in this war were the AI target generation systems developed by Unit 8200 of Israeli military intelligence. Criticism that they collapsed ethical boundaries and overwhelming operational efficiency. These systems proved both at once.

Lavender uses semi-supervised machine learning to score people. The target is the entire population of Gaza and Iran-linked proxy forces. It assigns each person a number indicating the likelihood of belonging to an armed group. Call patterns, social media activity, movement routes, frequency of phone-number changes, contact with numbers linked to armed organizations. By combining dozens of kinds of surveillance data, Lavender marked 37,000 Palestinians as "suspected militants" in the first weeks of the war.

The April 2024 investigation by +972 Magazine and Local Call, based on testimony from six Israeli intelligence officers, was shocking. Officer "B" said this: "I invested 20 seconds for each target and handled dozens a day. As a human, I had no added value beyond serving as an approval stamp." Verification meant only confirming whether the target was male. Lavender's error rate was acknowledged to be about 10%, but the military accepted it and approved use. The calculation is plain. It is 10% of 37,000 people. About 3,700 people were wrongly designated as targets. According to a source, "there was no supervision mechanism for detecting civilian men marked by mistake."

Gospel selects buildings, not people. Before its introduction, human analysts produced about 50 targets per year. Gospel generates more than 100 per day. From 50 per year to 100 per day. In the first 35 days of the 2023 war, more than 12,000 targets were bombed. That was twice the 6,000 struck during the 51-day war in 2014. A former Israeli intelligence official described the system as a "mass assassination factory." Gospel was also used to generate "shock targets," such as residential high-rises and public buildings, to apply psychological pressure to Palestinian society.

Where's Daddy? tracks targets marked by Lavender in real time. When the person enters the family home, it automatically sends an alert. One intelligence officer testified: "We were not interested only in killing Hamas operatives when they were in military buildings. The army bombed them at home without hesitation. Bombing a house was much easier." Lower-grade targets were struck with unguided "dumb bombs." Entire homes collapsed. Everyone inside died. For one low-ranking combatant, the deaths of 15-20 civilians were permitted. For senior commanders, the deaths of more than 100 civilians were approved several times.

In December 2025, the IDF created Bina, a dedicated AI division. The goal of this unit, led by a brigadier general, was clear: "to turn one tank into 100, and one soldier into 100." In April 2025, Genie, a military AI chatbot modeled on ChatGPT, was deployed to all military command centers. Using RAG technology, it answers commanders' questions based on IDF operational data. Fire Weaver networked battlefield sensors and weapons. An algorithm allocated in real time which target should be struck by whom, when, and how. This is the background that allowed Israel to strike Iran's leadership, air defenses, and missile bases almost simultaneously.

Action on Armed Violence, AOAV, assessed that this entire system reduced human analysts to "biological rubber stamps." One intelligence source described what he observed while accompanying ground forces: "People are promoted to terrorist rank after they die."

Chapter 5. The 20-Second Stamp, the Ethical Abyss Between AI and Civilian Massacre

There is a most painful question in this war. AI promised accuracy. Civilians must be protected. The gap lies between those two statements.

As of December 2025, the death toll in the Gaza war had exceeded 70,117. More than 170,000 people were wounded. A leaked Israeli military intelligence database reported by The Guardian and +972 in August 2025 showed the reality. By May 2025, confirmed Hamas/PIJ combatant deaths were about 8,900. That was about 17% of the 53,000 total deaths at the time. The remaining 83% were civilians. The Max Planck Institute estimated total deaths at 100,000-126,000. 27% were children under 15. 24% were women.

In Operation Epic Fury as well, errors in AI targeting produced deadly results.

An elementary school in Minab, Iran, was mistakenly bombed, killing about 100 children. An airstrike on a girls' school in Hormozgan Province killed 165 people. Public sources cannot determine whether this was due to AI recommendations, data bias, or traditional military error. But when decisions are compressed and recommendations are automated, the actual review time shrinks. The risk of civilian harm grows. This concern is structural.

The criticism from international law experts focuses on three principles.

First is the principle of distinction. Lavender's error rate is 10%. That means thousands of civilians were classified as combatants. Can a probability score based on digital patterns satisfy the positive identification required by international humanitarian law?

Second is the principle of proportionality. The deaths of 15-20 civilians were allowed for one low-ranking combatant. In previous conflicts, the number was 0-5. Automated systems cannot perform proportionality assessments, which require value judgments. Responsibility for civilian harm becomes unclear. A "responsibility gap" appears.

Third is the duty to take precautions in attack. A 20-second human review is assessed as "almost irrelevant" to meaningful legal review.

UN Special Rapporteur Francesca Albanese concluded that Israel was committing genocide. In November 2024, the ICC issued war-crimes arrest warrants for Netanyahu and Gallant. The president of the International Committee of the Red Cross repeatedly urged compliance with the laws of war and recommended banning unpredictable autonomous weapon systems and autonomous weapons that directly target humans.

Chapter 6. Overwhelm Quality with Quantity, Iran's Asymmetric AI Strategy

Iran's military philosophy starts from different premises than that of the United States and Israel. Instead of developing cutting-edge AI, it makes large numbers of cheap weapons and economically exhausts expensive defense systems. It is a doctrine of asymmetric attrition.

The cost contrast shows the core of this strategy. Iran spends $20,000-50,000 on one Shahed drone. The defender spends $200,000-280,000 to intercept it. Attack 1, defense 20. Iran's drone production capacity is 200-500 units per month, 2,400-6,000 per year. At Russia's Alabuga factory, Shahed-type drones were also being produced at a level of 18,540 per year.

In February 2025, Iran commissioned its first drone aircraft carrier, IRIS Shahid Bagheri. It placed a 570-foot runway on a converted 787-foot container ship. The drones stockpiled in a vast underground tunnel network were dispersed assets that no airstrike could completely eliminate.

The most technically notable weapon is the Fattah-2 hypersonic missile. It carries Iran's first true hypersonic glide vehicle. Its speed is Mach 15. Inside the atmosphere, it can maneuver in both pitch and yaw. On March 1, 2026, the Fattah-2 was used in combat for the first time. It penetrated Israel's multilayer air defenses and struck an IDF command center, killing seven senior officers. Rafael vice president Yuval Baseski warned, "Hypersonic missiles have opened a new era of air defense. We cannot rely on traditional approaches."

Iran's drone "swarms" were not AI autonomous swarms in the Western sense. They were mass coordinated attacks poured out in waves along preprogrammed GPS waypoints. But the scale was different. In Operation True Promise IV in 2026, Iran launched more than 2,000 suicide drones across nine countries within a few days. It mixed drones with ballistic missiles to exhaust the defender's radars and interceptor inventories at the same time. This was the result of closely observing Russia's night swarm tactics of more than 100 drones in Ukraine and applying them to the Middle East.

Meanwhile, the United States used Iran's strategy in reverse. It reverse-engineered Iran's Shahed-136 and created LUCAS, a low-cost autonomous attack drone priced at $35,000 per unit. Hundreds of LUCAS drones operated by Task Force Scorpion Strike formed swarms and saturated Iran's air-defense network. They precision-struck radar bases and missile launchers. A case was reported in which one $30,000 drone destroyed a $300 million radar site. One drone against one radar. The cost difference is 10,000-fold. The United States turned Iran's own asymmetric logic back on it.

Iran's domestic AI development is often described as "big thunder, no rain." It established a $20 billion national AI investment plan and a National Artificial Intelligence Organization, but sanctions have limited access to advanced chips and cloud infrastructure. In May 2025, however, there was a finding that could not be ignored. Wreckage of Iranian drones recovered in Ukraine confirmed the first real-world deployment of a fully autonomous lethal system capable of AI-based target selection without human input. The drone itself may have been crude. But the seed of autonomy planted inside it was enough to make Western analysts tense.

Chapter 7. The 2.4-Second Judgment: From Iron Dome to Arrow 4

Israel's layered air defense system shows the most widely accepted form of military AI use. It is a defensive system that operates at a speed where humans cannot intervene.

Iron Dome's AI takes 2.4 seconds from detection to engagement. The EL/M-2084 radar and mPrest battle management system analyze 18 variables, including altitude decay rate and wind shear patterns, to calculate the optimal intercept moment. It ranks threats by population density and strategic value. About 70% of incoming projectiles are judged harmless and ignored. That saves interceptor missiles. No human can analyze 18 variables and make a judgment in 2.4 seconds.

After October 7, 2023, Iron Dome received at least five major software upgrades. In March 2025, a special upgrade was made to counter drones and cruise missiles. Its engagement range expanded from its early ability to intercept only short-range rockets to cruise missiles, ballistic targets, UAVs, and precision-guided weapons. On December 30, 2025, the laser-based Iron Beam system entered operation. It was reported to have been used in combat for the first time in March 2026 against Hezbollah rockets launched from Lebanon.

David's Sling achieved the first ballistic missile interception in its history during the Iran war in June 2025. The most significant development is the arrival of Arrow 4. The IAI CEO officially announced it in February 2026. It was designed with AI-based technology to intercept maneuverable reentry vehicles (MaRV) and multiple-warhead missiles (MIRV). It was a direct response to Iran's attempt to introduce multiple-warhead missiles and saturate the air defense network. Israel added $12.5 billion to its 2025-2026 defense budget. It tripled the production rate of Arrow 3.

AI autonomy in defense and AI autonomy in attack stand on entirely different moral ground. Intercepting an incoming missile within 2.4 seconds. Few people oppose that. But when that very capability, speed, autonomy, and the exclusion of humans, moves to the offensive side, the moral debate shifts to a completely different level.

Chapter 8. War Faster Than the Speed of Thought

In World War II, it took six months. To gather intelligence and carry out a bombing.

In Iraq, 2,000 people were involved. To identify one target.

In Iran, 20 people finished it in a matter of hours.

The Guardian described the speed of this operation as "faster than the speed of thought."

This is the central paradox of AI war. If machines wage war at a speed beyond human judgment, what does human control mean?

Paul Scharre, a former Pentagon official for autonomous weapons policy, warns: "As militaries integrate AI and autonomy more deeply, the speed of combat action could reach a threshold too fast for humans to respond." A West Point hypothetical scenario shows this. An AI-equipped MQ-9 Reaper drone detects enemy forces. It gives the operator 15 seconds. The operator is still hesitating at the third second. The drone does not wait. It engages autonomously and kills six noncombatants.

Automation bias is a more serious problem. Research shows that people tend to comply with decisions made by computers over what they see and hear with their own eyes and ears. In war simulations, AI models escalated to nuclear options in 95% of Cold War-style nuclear crisis scenarios. Choosing de-escalation was a rare exception. Chinese military analyst Chen Hanghui has projected a "battlefield singularity." It is the point where machines make decisions and war proceeds at a speed humans cannot keep up with.

U.S. Secretary of Defense Hegseth announced an "AI-first combat power" strategy in January 2026. Frontier AI models are to be deployed to soldiers within 30 days of public release. Demonstrations of autonomous drone swarms were made mandatory. In a world where 900 airstrikes can be carried out in the first 12 hours, the time for diplomats to make phone calls is evaporating.

Chapter 9. Gunpowder, Nuclear Weapons, and AI: The Third Revolution in Warfare

In Libya in 2020, Turkey's Kargu-2 drone carried out the first autonomous engagement against humans. In Nagorno-Karabakh, drones destroyed 45% of all targets. In Gaza in 2021, Israel declared the "first AI war." In Ukraine, both sides experimented with large-scale drone warfare and AI navigation. In each conflict, AI's autonomy, speed, and centrality rose step by step.

In Libya, one drone fought on its own.

In Nagorno-Karabakh, drones decided victory.

In Ukraine, both sides pushed the front lines with thousands of drones.

In Gaza, AI made lists of people to kill.

In the 12-Day War of 2025, AI was integrated across all domains.

In the 2026 Iran operation, large language models were inserted into the kill chain. It was the first full-scale war of its kind.

In 2015, more than 3,000 experts, including Stephen Hawking and Elon Musk, warned that lethal autonomous weapon systems could trigger "the third revolution in warfare, comparable to gunpowder or nuclear weapons." The UN Secretary-General and the president of the ICRC called for a legally binding treaty on autonomous weapons by 2026. That goal was not achieved. Discussions in Geneva on rules for lethal autonomous weapons have stalled because of disagreements among major powers.

The warning from Kissinger and Allison echoes: "No great power in history has ever refrained from developing a technology of its own while fearing that a rival could apply that technology to threaten its survival and security."