The First
AI-Integrated
War

The war unfolding across Iranian airspace is unlike anything that has come before it. Not simply because of its scale — though the scale is staggering — but because of what it reveals about the world we have built: a world in which artificial intelligence processes targeting data in classified networks, where a $20,000 drone can neutralise a $2 million interceptor, where the most significant ethical confrontation of the conflict was fought not between armies but between a Silicon Valley company and the United States Secretary of Defense. What began as a decades-long standoff over nuclear centrifuges has metastasised into a live field test of 21st century warfare doctrine. The results will shape military strategy, international law, and AI governance for generations.

Part I — The Escalation: From Proxy War to Decapitation Strike

To understand where we are, you need to understand how we got here. After the broader Middle Eastern crisis began in 2023 and Iran and Israel exchanged missile strikes in 2024, Israel and the United States launched the Twelve-Day War in June 2025 — a defined, targeted campaign against Iran's nuclear infrastructure. That conflict struck the facilities at Natanz, Fordow, and Isfahan. It killed nuclear scientists and IRGC commanders. It ended, after twelve days, with a fragile US-brokered ceasefire brokered through Oman.

In the opening salvo, the Israeli Air Force launched five waves of airstrikes using more than 200 fighter jets to drop more than 330 munitions on roughly 100 targets. Iran retaliated with over 550 ballistic missiles and more than 1,000 suicide drones. The United States entered on June 22 with bunker-buster strikes on the hardened nuclear facilities. Trump declared Iran's nuclear capabilities neutralised. It was not the end. It was the intermission.

In January 2026, Iranian security forces killed thousands of protesters in their crackdown on the largest nationwide uprising since the 1979 revolution. Even as domestic instability grew, Iran quietly reconstituted — importing chemical precursors for solid rocket propellant, rebuilding missile production lines, storing highly enriched uranium in underground facilities the IAEA had not been permitted to inspect since the June strikes. Iran's foreign minister declared a historic agreement was "within reach" as late as February 25, 2026. Three days later, the United States and Israel launched a fundamentally different kind of offensive. The opening salvo assassinated Supreme Leader Ayatollah Ali Khamenei and several family members in Tehran.

The codenames — Israel's "Operation Roaring Lion" and America's "Operation Epic Fury" — capture something real about the ambition. The IDF has struck over 1,700 assets of Iran's military industry, claimed to have destroyed or disabled roughly 70 percent of Iran's estimated 500 ballistic missile launchers, and achieved air supremacy over most of Iran's airspace after taking out more than 100 air defense systems. US military strikes have hit over 8,000 targets across Iran, including 130 enemy vessels. Iran's response has been to widen the map. Tehran has launched strikes across nine countries, targeting US military presence in all Gulf states while restricting traffic through the Strait of Hormuz and driving oil prices past $100 per barrel.

Part II — The Drone Economy: How Iran Rewrote the Rules of Attrition

The most consequential military technology of this conflict is also among the cheapest: the one-way attack drone. Iran's Shahed-136 — constructed largely from foam, plywood, and a 50-horsepower piston engine with a distinctive moped-like sound — has proven to be the great equaliser of this war. In the first week of fighting alone, Tehran fired nearly 2,000 drones at US bases and allied targets across twelve countries. Gulf allies were running critically low on air defense interceptors within just four days of the conflict starting.

The arithmetic is brutal. Iran fires a $20,000 drone. The defender responds with a Patriot interceptor costing hundreds of thousands. Do that exchange a thousand times and the defender has spent a billion dollars stopping twenty million dollars' worth of threat. The first week showed that 2,000 Shaheds are extremely difficult to intercept and can cause serious damage. Ukraine has demonstrated interception rates of roughly 80 percent against individual Shahed strikes — but at larger volumes, even the most capable defenses can be overwhelmed, leaving hundreds of munitions free to seek their targets.

The newest Shahed variants are significantly more capable than the originals. Ukraine has found AI-enabling Nvidia chipsets in Shahed wreckages, as well as thermal-vision modules capable of locking onto targets at night — and the drones are now interconnected, allowing them to exchange information with each other mid-flight. What began as a pre-programmed GPS follower has evolved, through years of battlefield iteration, into something approaching cooperative autonomy.

Iran has also debuted an entirely new weapons class. On March 1, 2026, an Iranian unmanned surface vessel struck a Marshall Islands-flagged oil tanker in the Gulf of Oman — the first confirmed state-led deployment of explosive drone boats against commercial shipping. One drone technology CEO described the threat geometry: a single operator can control a swarm of ten autonomous boats acting with significant independence, pre-programmed for coordinated attacks. The same saturation logic that makes aerial drone swarms so difficult to stop now applies to the surface of the water.

The counter-drone picture is equally remarkable. The US rushed 10,000 Ukrainian-developed Merops interceptor drones to the Middle East — AI-enabled systems costing roughly $14,000 each, cheaper than the Shaheds they are designed to kill, stress-tested in Ukraine's four years of nightly drone warfare. The Pentagon says Iranian drone attacks are now down 95 percent from their peak. Ukraine, which produced approximately 4.5 million drones in 2025 alone, has become the world's foremost authority on stopping Shaheds — and US forces formally requested Ukrainian specialists to deploy to the Gulf. The most tactically valuable expertise in this theatre came not from West Point or Sandhurst, but from Ukrainian drone engineers.

Part III — Iron Beam: The Laser Enters Combat

One of the most consequential technological milestones of this conflict is the operational debut of directed-energy air defense — a capability the world has been developing toward for decades and which Israel became the first nation to field at operational scale.

On December 28, 2025 — just two months before the war began — the Israeli Ministry of Defense delivered the first operational Iron Beam high-power laser interception system to the IDF. It was, officials noted, the first time in human history that a high-power laser interception system had reached full operational maturity anywhere on Earth. The Iron Beam operates in the 100–150 kilowatt power class, using advanced beam-control optics to focus energy on targets at tactical ranges of up to 10 kilometers. It burns through rocket casings and drone structures within seconds of target acquisition. Its decisive strategic advantage is cost: each interception requires only a few dollars' worth of electricity, compared to tens of thousands for a single Iron Dome Tamir missile.

That cost ratio matters enormously against the saturation warfare Iran has perfected. While existing kinetic systems achieve interception rates of 90–95 percent against conventional missiles, their performance against drones has historically been closer to 50 percent. Iron Beam is specifically designed to counter very low-flying devices tracking the ground — a capability that previous systems lacked. The system completes a five-layer defense architecture, handling the short-range, low-cost threats that kinetic systems cannot economically engage.

Rafael has additionally developed an Iron Beam-M — a mobile 50-kilowatt variant for five-to-seven kilometer protection — and the Lite Beam, a 10-kilowatt system mountable on tactical vehicles. Both are already deployed and have been used in active scenarios. The fundamental physics of Iron Beam — firing at the speed of light from an effectively bottomless magazine — represents something qualitatively different from every previous air defense system. It is the end of the missile-versus-missile exchange equation that has governed air defense economics since the Cold War.

Part IV — Space, Cyber, and the AI Command Layer

The kinetic dimension of this conflict sits atop an invisible infrastructure of satellites, digital networks, and machine intelligence that enables everything else. The US Space Force has emerged as a critical provider of real-time missile warning data across the region. Orbital sensors detect the infrared heat signatures of Iranian ballistic missile launches within milliseconds, allowing automated defense systems to calculate interception trajectories before most human operators could process the alert. Commercial satellite networks — particularly SpaceX's Starlink and Starshield low-Earth-orbit constellations — provide high-bandwidth, unjammable communications enabling continuous control of autonomous drone swarms even in heavily contested electronic environments.

The cyber dimension was opened before the first missile fired. In the opening hours of the war, IDF and US Cyber Command carried out disabling operations against Iranian military telecommunications networks, delaying and disrupting Iranian counter-offensives. Israeli cyber units had also gained long-term access to Tehran's municipal traffic cameras months before the war began, allowing analysts to track the daily movements of targeted Iranian officials — the pattern-of-life data that made the opening decapitation strike possible.

This conflict is broadly being characterised as the world's first truly AI-integrated war. AI is operating simultaneously across multiple layers: processing satellite imagery and signals data to identify targets, guiding autonomous drones through GPS-denied environments, and coordinating real-time logistics and munitions allocation. According to the Washington Post, Claude — Anthropic's AI model, deployed via Palantir's Maven Smart System on classified networks — was generating proposed targets in the Iran campaign. That last detail is the hinge upon which the next part of this story turns.

Part V — The Anthropic Standoff: When an AI Company Said No

In July 2025, the Pentagon awarded $200 million contracts to four AI companies — Anthropic, OpenAI, Google, and Elon Musk's xAI — to "prototype frontier AI capabilities that advance US national security." Anthropic's partnership with Palantir made Claude the only AI model used in classified missions, deployed across multiple US national security agencies through its "Claude Gov" model. The company had made its conditions clear from the outset: it did not want its technology used for mass surveillance of people in the United States, or for fully autonomous weapons systems. Those restrictions were not abstract philosophical preferences — they were embedded in the contract language itself.

The confrontation came in January 2026. Defense Secretary Pete Hegseth issued an AI strategy memo instructing that all DoD AI contracts must, within 180 days, include standard "any lawful use" language — effectively requiring the removal of any vendor restrictions beyond what is required by law. The Pentagon CTO described what he found when he reviewed existing contracts: "dozens of restrictions, and yet these AI models were baked into some of the most sensitive and important places in the US military, where we do exercise combat power."

Anthropic's response was unequivocal. CEO Dario Amodei announced that the company would not concede, writing that Anthropic could not "in good conscience" grant the Pentagon's request. The company identified two specific objections: it did not believe current frontier AI models are reliable enough for use in fully autonomous weapons — and it regarded mass domestic surveillance of Americans as a fundamental violation of rights. Anthropic added that the Pentagon's latest "language framed as compromise was paired with legalese that would allow those safeguards to be disregarded at will."

At 3:47 PM on February 28 — before the deadline had even arrived — Trump posted on Truth Social calling Anthropic "A RADICAL LEFT, WOKE COMPANY" and directed every federal agency to immediately cease all use of its technology. Hegseth followed through on the supply chain risk designation, meaning any company that does business with the US military would have to prove they have no contact with anything related to Anthropic. The designation threatened far more than the $200 million contract — much of Anthropic's commercial success stems from enterprise contracts with large companies many of which may have Pentagon contracts or aspirations to them.

Anthropic, valued at roughly $380 billion, sued. On March 26, 2026, a federal judge issued a temporary injunction against the DoD, agreeing that their actions appeared to be "classic First Amendment retaliation" — and describing the Pentagon's stance as "Orwellian." The case, and a companion action in Washington DC, will likely reshape how commercial AI systems are used by the US military for years to come.

Part VI — OpenAI Steps In: Same Red Lines, Different Architecture

Within hours of Trump's ban, a rival company stepped into the vacuum. OpenAI CEO Sam Altman announced that the company had struck a deal with the Pentagon for its AI tools to be used in the military's classified systems — with, apparently, similar restrictions to those Anthropic had just been punished for insisting upon. OpenAI published its agreement structure, outlining three red lines: no use for mass domestic surveillance, no use to direct autonomous weapons systems, and no use for high-stakes automated decisions such as social credit scoring.

The announcement initially left observers confused. Had Altman somehow convinced Hegseth to accept the exact terms he had just denied to Amodei? The answer, on close examination, is more subtle and more troubling. The key distinction between the Anthropic and OpenAI approaches is largely a matter of framing — and framing determines who has interpretive authority. Anthropic had sought explicit contractual prohibitions with the vendor retaining the power to enforce those limits. OpenAI's agreement defines similar guardrails by reference to existing legal authorities and Defense Department policy, with interpretive discretion resting with the government. OpenAI, in other words, deferred to the Pentagon's own interpretation of what was lawful. The Electronic Frontier Foundation put it plainly: OpenAI was rightfully facing widespread criticism for filling the gap the DoD created when Anthropic refused.

By March 2, Altman posted amended contract language explicitly stating that the AI system "shall not be intentionally used for domestic surveillance." Critics noted that "consistent with applicable laws" is not an independent prohibition — it refers to authorities the Pentagon itself interprets and applies. The enforceability of OpenAI's red lines ultimately depends on the goodwill of the very institution being constrained. OpenAI's original 2023 policy had explicitly banned military use entirely. The company gradually softened this stance through 2024–2025. CEO Sam Altman acknowledged the optics: "This was definitely rushed. The optics don't look good." He defended the decision as necessary, arguing that adversaries — particularly China — are integrating AI into military systems without any safeguards whatsoever. That argument carries weight. The question is whether it also carries accountability.

Part VII — The Existential Risk Landscape

This is where the analysis must stop being tactical and become civilisational.

The nuclear question. Before the June 2025 strikes, Iran's nuclear breakout time — the period required to produce enough fissile material for a nuclear weapon — was estimated at nearly zero. The IAEA considered Iran to have enough nuclear material for nine nuclear weapons if further enriched to weapons grade. Whether the subsequent strikes substantially altered that calculus remains genuinely unknown. Military force can destroy infrastructure, but Iran will retain the nuclear expertise and likely key materials necessary for building a bomb at the end of this conflict. The IAEA has been shut out entirely since June 2025. No inspectors. No verified inventory of remaining stockpiles. The international community is, in the most literal sense, flying blind.

Leadership succession and hardening. With Khamenei dead, his son Mojtaba — reportedly more hardline than his father — has been installed as Supreme Leader. Khamenei senior reportedly blocked multiple internal pushes from the Revolutionary Guards to cross the nuclear threshold. Persistent rumours within the Revolutionary Guards suggest North Korea may be cooperating on compact warhead design — potentially shortcutting Iran's weapons timeline from years to months. Whether Mojtaba holds the line under existential pressure is a question no analyst can answer with confidence.

The proliferation cascade. Iran was attacked during two active rounds of negotiations — June 2025 and again in late February 2026. For states weighing whether to engage diplomatically with Washington, that precedent is significant. Dialogue did not protect Iran. Voices within the Iranian regime are now arguing that Tehran should withdraw from the NPT entirely and build the bomb. Countries including South Korea and Saudi Arabia are likely to move significantly closer to developing the technical means — and the political motivation — to build their own. Saudi Crown Prince Mohammed bin Salman has repeatedly stated the kingdom would seek a nuclear weapon if Iran acquires one. The logic of deterrence, once activated, tends to propagate.

The Strait of Hormuz as a civilisational chokepoint. Roughly 20 percent of the world's oil transits the Strait of Hormuz. Iran has already partially restricted traffic, with Brent crude spiking above $100 per barrel. A sustained closure — or strikes on the Kharg Island oil export terminal, which Iran has explicitly threatened — would constitute an energy shock with cascading consequences well beyond the Middle East.

The AI accountability void. Multiple analysts have noted that this conflict is outpacing global rules of war, with no established international framework governing AI-assisted targeting, autonomous drone swarms, or accountability when machines help determine who lives and who dies. The Anthropic-Pentagon standoff made this concrete: the question of what values to embed in military AI is too important to be resolved by contract disputes between a tech company and a government. It requires legislation. Congress has not provided it.

What This War Is Actually Teaching Us

Step back from the immediate violence and the technology reveals something deeper about the era we have entered. The Iran conflict is the first war in which every major domain of modern military capability has been deployed simultaneously and at scale: precision mass drone saturation, directed-energy interception, space-integrated real-time battle management, AI-assisted targeting, autonomous naval systems, and pervasive cyber operations — all in the same theatre, against the same adversary, at the same time.

The lesson from the Anthropic standoff is not that one company was brave and another was not. It is that the framework does not exist. The rules that should govern AI in warfare — rules with democratic legitimacy, statutory force, and independent enforcement — have not been written. What we have instead is regulation by contract: bilateral agreements between vendors and governments, enforced by whoever has more leverage. In the current case, that leverage belongs entirely to the government.

The cost asymmetry that defines this conflict — $20,000 Shaheds against million-dollar interceptors — will not disappear when the fighting ends. It will intensify. Every military in the world is studying what is happening in Iran right now. The conclusions they draw will shape procurement decisions, doctrine, and escalation thresholds for the next half century.

And underneath all of it — the drones and lasers and satellite networks and AI targeting systems and legal battles — sits the oldest and most dangerous technology in the human arsenal: uranium enriched to near-weapons grade, stored in underground bunkers that international inspectors can no longer access, in a country whose leadership has been violently decapitated and replaced by men with less reason to exercise restraint.

The existential risk is not theoretical. It is administrative. It is legal. It is political. It is the accumulated consequence of decades of decisions made by states, companies, and individuals who understood the risks and proceeded anyway. We are living in the world those decisions made.