Unveiling the first high-level assassination operation led by AI: How did Claude and Palantir kill Khamenei?

Source: Microsoft AI

Written by: Xia Chu

Today, a chilling stillness hangs over Shemiran, a district in northern Tehran. For Iran's Supreme Leader Ayatollah Ali Khamenei, this stillness usually signifies security, but on this day, it becomes a prelude to death.

This raid, codenamed "Operation Epic Fury," was not a traditional large-scale bombing, but rather a "surgical operation" woven from low-level code, real-time remote sensing algorithms, and distributed computing power. After the attack, Trump confirmed Khamenei's death on social media.

The significance of this operation lies in the fact that it was the first high-level decapitation strike in human history to be completely dominated by artificial intelligence (AI) within a "kill chain." In his command post deep underground in Tehran, Khamenei may have thought he had evaded satellites, but he didn't realize that he was facing not a single weapon, but a global surveillance and strike network comprised of Palantir, Anduril, and the top-tier large language model (Claude). This network no longer relied on expensive traditional platforms, but rather on "software-defined weapons."

According to the Wall Street Journal, in this war, AI is no longer an auxiliary tool; it has become the true decision-maker, tracker, and executor.

Palantir, Silicon Valley's "War Operating System"

Behind the scenes of the decapitation strike, Palantir's technology platform played the role of the "battlefield brain." Founded by Peter Thiel, the company's core mission has always been to break down data silos between intelligence agencies.

Breaking down the "ontology" of isolated islands

Palantir's most powerful weapons are its Artificial Intelligence Platform (AIP) and its flagship product, Gotham 5. In traditional command systems, intelligence analysts manually compare satellite imagery, communications snooping records, and open-source social media data. But in Operation Epic Fury, Palantir's Ontology technology transformed this chaotic data into intuitive, tangible objects.

The so-called "ontology" maps complex enterprise or battlefield data into easily understandable entities, such as "personnel," "locations," or "launch pads." By integrating data from ERP systems, sensors, satellites, and network monitoring into a "Common Operating Picture" (COP), commanders are no longer faced with dry reports, but with a real-time digital twin of the battlefield.

Frontline Deployment Engineers: Programmers on the Battlefield

To ensure this complex system operated in Tehran’s high-intensity electronic warfare environment, Palantir deployed a special group of warriors—Forward Deployed Engineers (FDEs). These engineers did not sit in air-conditioned rooms in Denver or Silicon Valley, but instead wore tactical vests and were directly embedded in the combat units of the U.S. Central Command (CENTCOM).

This wartime capability reduced system updates that would normally take months to just hours. When Khamenei was killed, it was the FDE that adjusted the MetaConstellation satellite scheduling logic in the background, ensuring that more than three satellites simultaneously performed cross-validation the instant the target left the bunker.

StarShield Unveiled: SpaceX's Super Battlefield Broadband

To understand this operation, one must first understand how the US military broke through Iran's impenetrable electromagnetic blockade.

Before the operation began, Tehran cut off terrestrial internet and mobile communications throughout the country in an attempt to render U.S. military sensors "blind." However, according to The Wall Street Journal, the U.S. military used SpaceX's most secretive asset— Starshield—and the MILNET satellite constellation behind it .

This is no longer the semi-civilian Starlink terminal seen on the Ukrainian battlefield. StarShield consists of approximately 480 dedicated hardened satellites, integrating the highest security encryption protocols at the NSA level. In the operational logs of the U.S. Central Command (CENTCOM), these satellites are figuratively referred to as "digital oxygen": when the Iranian Revolutionary Guard attempted to sever frontline communications using the Russian-made Kalinka jamming system, StarShield established a perpetually uninterrupted aerial grid in orbit via laser inter-satellite links reaching speeds of up to 200 Gbps.

Most formidable was the compact terminal known as the UAT-222 . Measuring only two feet square, it could be carried by a special forces soldier. When this tiny cube was activated in a bunker near the Tehran residence, petabytes of high-resolution images and electromagnetic signals that would normally take hours to transmit were transmitted directly into Palantir's analysis engine within seconds, penetrating the jamming smoke.

Claude: A Battle Over the Soul of AI

However, the process of AI killing Khamenei sparked a fierce conflict within the United States regarding AI ethics. The focus of this conflict was Claude, the top-tier AI model developed by Anthropic.

As the only state-of-the-art large-scale model authorized by the Pentagon to run on a highly classified, physically isolated network, Claude was once the most relied-upon tool for U.S. military intelligence analysts. Its "Claude Gov" version performed exceptionally well in processing massive amounts of intercepted Persian classified documents.

Claude's role in the operation was not to directly control weapons, but rather to process massive amounts of unstructured warfare data. According to declassified documents, the US military first used Claude on a large scale for "intelligence synthesis" in its operation against Venezuelan leader Maduro in early 2026. Claude was able to quickly read thousands of hours of intercepted Persian communications, identify cracks in the command chain within the Iranian Revolutionary Guard, and generate dozens of dynamic game-theoretic strike scenario simulations for commanders.

Analysts no longer need to write lengthy briefings; they can simply ask, as casually as ordering food: "If we were to electronically suppress Tehran at this moment, and simultaneously launch an air strike, what would be Khamenei's most likely escape route?" Claude would then provide an optimized interception probability chart based on its vast military theoretical training and real-time intelligence streams.

However, according to an exclusive report by The Wall Street Journal in February 2026, an open hostility erupted between the Trump administration and Anthropic CEO Dario Amodei. Defense Secretary Pete Hegseth demanded that Anthropic remove all of Claude's guardrails in order to integrate them directly into the fully automated lethal weapon system.

Instead, OpenAI and Elon Musk's xAI took center stage. xAI was quickly placed at the heart of the US military's most secretive missions because it promised computing power "unbound by political correctness." Ironically, in the operation against Khamenei, the Claude model running on the Palantir platform still played a crucial supporting role—although it refused to pull the trigger directly, it had cleared the fog of intelligence for the final strike by processing previous covert operations against Venezuelan President Maduro and petabytes of data intercepted in Tehran.

"Where's Dad?": Algorithms track everyone.

If Palantir and Claude provided strategic-level intelligence, then the three AI systems developed by the Israel Defense Forces (IDF) reveal the most chilling logic at the tactical level. These three systems are collectively known as the "mass assassination factory."

Lavender and Habsola

In its operations against Tehran, the U.S. military drew on algorithms honed by the IDF in combat operations in Gaza.

• "The Gospel": An AI system specifically designed to recommend targets for construction. It can generate strike lists at a rate of 100 per day, while humans have historically been able to generate only 50 per year.
• "Lavender": It scores millions of people and automatically tags suspected militants by analyzing social networks, movement patterns, and call records. At its peak, it tagged 37,000 targets.

The Fatal "20-Second" Decision

The most controversial aspect lies in the role of humans. According to The Guardian, after these AI systems recommend targets, human commanders often spend only "20 seconds" reviewing them. That 20 seconds is merely enough to confirm whether the target is male.

Even more brutal is a system called "Where's Daddy?". Instead of tracking aircraft like traditional radar, it tracks the association between targets and their home residences. The system automatically monitors when tagged individuals enter their homes. Commanders believe that launching an attack when these individuals are home with their families is easier than attacking military outposts, even though it means that the entire building's civilian population could become "collateral damage."

In the course of Khamenei's assassination attempt, this logic was elevated to the level of national leaders. The algorithm no longer searched for Khamenei's luxury cars, but rather for every minute detail about him.

Anduril and Shield AI: Software-Defined Air Dominance

In order to carry out the final strike, the U.S. military no longer uses expensive stealth aircraft repeatedly, but instead uses Cooperative Combat Aircraft (CCA) defined by new defense companies such as Anduril and Shield AI.

A key technological highlight of this operation was the drone swarm's ability to autonomously adjust its formation based on real-time threat perception after entering Tehran's airspace. When Iranian air defense radar locked onto one of the drones, the entire swarm shared this threat through the Lattice software system and automatically dispatched a group of drones for electronic guidance and anti-radiation strikes. This "software-driven" warfare renders traditional, hardware-centric defense systems cumbersome and outdated in the face of algorithmic iterations.

Shield AI focuses on developing what it calls "the world's best AI pilot"—Hivemind. This software allows unmanned systems to perform complex tasks even when completely without GPS, satellite communications, and human operators.

Hivemind's technological backbone is EdgeOS, a middleware environment specifically designed for high-performance real-time robotics. Its core features include:

In-flight brain transplant: The power of the A-GRA architecture

In February 2026, Anduril demonstrated an experiment that shocked the military: its YFQ-44A drone successfully switched between two completely different AI systems during flight. The first half of the flight was controlled by Shield AI's "Hivemind" software, which allowed the drone to autonomously avoid obstacles and form formations like a bird; the second half seamlessly switched to Anduril's "Lattice" system to perform the final target lock.

This "aerial brain swap" relies on a modular standard known as the "Government Reference Autonomous Architecture" (A-GRA). This means that if an adversary develops electronic jamming against a certain AI, the drone can instantly download and run another algorithm, much like updating an app on a mobile phone.

Hawkeye Headset: A Soldier's "Digital Teammate"

During ground operations, U.S. special forces soldiers wore the EagleEye mixed reality headset, jointly developed by Anduril and Meta (formerly Facebook).

This headset is no longer a bulky bulletproof helmet, but a holographic display system integrating all data from the Lattice network. Soldiers can directly see the skeletal posture of enemies, the outlines of obscured targets, and even see real-time footage transmitted from aerial drones.35 Palmer Luckey calls it a "digital teammate for soldiers," giving everyone fighting on the front lines a God's-eye view synchronized with the Pentagon.

"New Military Industry": How Venture Capital is Reshaping the Armory

Behind Khamenei's assassination was a hidden check.

For decades, the arms trade has been the domain of traditional giants like Lockheed Martin. But now, Silicon Valley venture capital, through its "American Dynamism" strategy, has officially taken over the pace of research and development on the battlefield.

The "New Military-Industrial Complex" on Shanshan Road

Venture capital firms led by Andreessen Horowitz (a16z) completed a record $15 billion funding round in 2026. Their bets are no longer on food delivery or social media apps, but on hard tech companies like Anduril, Shield AI, and Saronic.

These companies operate in a completely different way from traditional contractors:

1. Speed: Traditional giants take ten years to develop a radar system, while these companies can do it in just a few months through software simulation.

2. Consumability: Instead of building a $100 million F-35, they aim to produce 10,000 autonomous drones worth $10,000 each.

3. Software first: In their eyes, weapons are just "code in an aluminum shell".

This shift in capital has given the United States a very high margin for error in its actions against Iran. Even if some drones are intercepted, the remaining drones can automatically fill in through the distributed Lattice network.

Three Clocks: The Strategic Limitations of AI Warfare

In the wake of Khamenei's death, military strategists began to reflect on the cost of this victory. They proposed the famous "three clocks" theory to examine conflicts in the age of AI.

1. Military clock: AI has drastically shortened the "sensor-to-shooter" time. Decapitation strikes that previously required months of preparation can now be executed within seconds of the algorithm confirming the target. The military clock has been sped up to its extreme.

2. Economic Clock: While AI weapons are inexpensive to produce, their rapid depletion puts exponential pressure on supply chains. If the conflict drags on, energy premiums, shipping risks, and inflation will backfire on the attacking economy.

3. The Political Clock: This is the slowest clock. AI can precisely eliminate a leader, but it cannot automatically win the consent of the local people or quell regional anger.

Khamenei's death proved the unparalleled dominance of algorithms in the "Find, Fix, Finish" cycle. However, when war becomes as casualty-free and efficient as clicking a screen, the political threshold for humans to initiate war is dangerously lowered.

The End and the Beginning of an Era: Software-Defined Geopolitics

This is the real process of how artificial intelligence killed Khamenei: there was no smoke-filled battlefield, no fierce aerial combat, only the constantly jumping data bars on the Palantir platform, the intelligence summary output by the Claude model, and the red outline drawn on the HUD by the Anduril Lattice system.

Khamenei's downfall marks the full-scale beginning of the era of "software-defined geopolitics".

As the Wall Street Journal commentary pointed out: We have entered a battlefield where human commanders don't even have time to feel fear.

So who is the winner?