AI-Waged War: The “Silent Quds” Operation That Redefined Modern Warfare
On March 1, 2026, news broke around the world that Iran’s supreme leader Ayatollah Ali Khamenei had been killed in a targeted strike. What makes this event historic is not just the political earthquake it sent across the Middle East, but how it was carried out: this was not a traditional military operation driven by human commanders step-by-step, but a war campaign largely designed, coordinated, and executed by artificial intelligence. In just 11 minutes and 23 seconds, the United States completed a precision assassination with zero American casualties and zero collateral damage—a milestone that marks AI’s first full-scale, high-stakes entry into real-world combat.
Codenamed “Silent Quds”, the operation seemed surprisingly small on the surface: only one MQ-9B SeaGuardian drone, an 8-member special forces advance team, and no large-scale deployment of troops or heavy weapons. The real striking power came from a tightly integrated AI war machine: Claude 4 Opus (Gov version) large language model, Palantir Foundry defense platform, and the JADC2 (Joint All-Domain Command and Control) system connected to U.S. Central Command (CENTCOM). This combination turned a risky, complex assassination into a streamlined, algorithm-driven mission.
The AI “Brain” Powering the Strike
Two systems formed the backbone of the operation, each handling a critical piece of the war machine.
First, Claude 4 Opus (Gov Edition). It is the only large language model authorized by the Pentagon to run on classified networks, purpose-built for military scenarios. With a 1M-token context window, multimodal processing, and an ASL-3 security rating, it excels at unstructured data analysis, multilingual comprehension, and combat simulation. In this mission, it achieved a 98.7% accuracy rate in real-time Farsi conversation transcription, analyzed encrypted Iranian Revolutionary Guard documents 82 times faster than human analysts, and simulated combat scenarios with an error rate of just 2.3%. Claude acted as the “thinking brain” of the entire operation.
Second, Palantir Gotham/Foundry. Backed by CIA investment since 2005, Palantir specializes in breaking down data silos using dynamic ontology and knowledge graph technology. It fused satellite imagery, electronic signals, human intelligence, and social media data to map the relationships, movement patterns, and security routines of Iran’s top leadership. Its MetaConstellation platform coordinated hundreds of satellites for 24/7 surveillance, detecting even subtle underground activity. Palantir served as the “central nervous system”, turning raw data into actionable targets and commands.
When combined, Claude and Palantir created a closed loop: data collection → analysis → decision-making → command execution. The old, layered military structure—where intelligence, command, and combat units passed information slowly and separately—was effectively erased.
The Six-Step AI Kill Chain: Perfected in Minutes
What truly astonished observers was the full AI-driven kill chain, which compressed hours or even days of work into minutes. Every stage was data-led, algorithm-dominated, with humans only giving final authorization.
1. Intelligence Awareness
Claude processed over 2.3 petabytes of data, covering 120 million intelligence fragments across space, air, land, sea, and cyberspace. A human team of 328 analysts would have needed 100 days to finish the same work; Claude did it in 90 minutes. Palantir’s data backbone supported 1.2 million heterogeneous data processes per second, laying the foundation for target identification.
2. Target Lock
Khamenei’s security relied on random location changes, multilayer guards, and full-band electromagnetic shielding, creating a 5-kilometer uncertainty radius. AI changed that.
Claude modeled six months of movement, schedules, guard shifts, and external variables like religion holidays and weather, generating over 1,000 possible routes and narrowing down the action window with 98.7% accuracy. Palantir then overlapped urban geography and air defense data to identify an 800-meter air-defense blind spot. The target’s uncertainty radius shrank from 5 kilometers to 500 meters, leaving a precise 3-minute window for attack.
3. Decision & Simulation
In just 8 minutes, Claude designed 15 complete raid plans, each including drone routes, missile parameters, electronic warfare timing, and evacuation routes. It simulated one plan every 30 seconds, scoring each for success rate, casualties, and collateral damage. The top plan scored 98.2/100: use stealth MQ-9B for low-altitude penetration, launch AGM-114 Hellfire missiles, and use EA-18G Growlers to jam enemy air defenses. Tactical data synchronized across units in under 200 milliseconds.
4. Cross-Domain Coordination
Traditionally, different military branches use incompatible communication systems, causing delays of minutes or longer. Claude translated tactical commands into a unified natural language, and Palantir ensured secure transmission. Within 3 seconds of the CENTCOM “execute” order:
- 0.5s: Order sent to drone, electronic warfare plane, and special forces
- 1s: EA-18G started full-spectrum jamming
- 2s: MQ-9B dove to 50 meters for stealth approach
- 3s: Special forces took blocking positions
Perfect synchronization, no human middlemen.
5. Precision Strike
The Hellfire missiles used AI terminal guidance, recognizing the target vehicle mid-flight and correcting trajectory to within 1 meter. Claude simulated blast power, shockwave range, and fragment spread to limit damage within 10 meters. Despite civilian buildings nearby, no bystanders were hurt, no collateral damage occurred.
6. Battle Damage Assessment
Claude used visual recognition to confirm the target was destroyed and no civilians harmed in 0.3 seconds—about 1,000 times faster than manual assessment. The mission was declared complete almost instantly.
From start to finish, the entire chain took only 11 minutes 23 seconds. Modern warfare had been rewritten.
Warfare Enters the “Data Overload” Era
This operation proved a brutal new reality: warfare has moved beyond human cognitive limits. We have entered the data overload era, where no team of analysts, commanders, or soldiers can process information fast enough to compete with AI.
Without AI:
- You cannot exploit the information advantage
- You cannot compress the kill chain to react in seconds
- You cannot achieve seamless joint operations across domains
In short, AI is no longer a “nice-to-have” — it is a necessity for survival on the modern battlefield.
Three Defining Trends of Future Intelligent Warfare
The “Silent Quds” operation is not an isolated trick. It reveals three clear trends that will shape global conflict for decades.
1. Algorithm Over Firepower
Future wars will be won by engineers as much as soldiers. The U.S., China, and Russia are already racing:
- U.S.: autonomous decision-making algorithms
- China: drone swarm cooperative algorithms
- Russia: electronic warfare anti-AI algorithms
The global military AI market reached $11.6 billion in 2025 and is projected to exceed $35.54 billion by 2031, growing at 14.49% annually. Weapons become just tools; algorithms decide who strikes first and wins.
2. Unmanned Battlefields
Human troops will increasingly be replaced by unmanned systems. The global military drone market surpassed $30 billion in 2025. The U.S. CCA program plans to equip every F-35 with 2–3 drone wingmen, with 60% of its combat platforms unmanned by 2030. War will be fought in the air, sea, and land by machines commanded by AI.
3. Low-Cost, Low-Casualty Intervention
This strike cost the U.S. about $10 million, just one-fifth the price of a traditional targeted operation. With zero casualties and minimal political backlash at home, major powers face far fewer barriers to launching unilateral strikes. The result? More small, sharp, AI-led conflicts, and a more volatile geopolitical world.
Closing Thought: War Becomes a “Game of Creation”
The article ends by quoting a chilling line from the ancient Chinese divination text Tui Bei Tu:
“Those who fly are not birds; those who dive are not fish. War does not depend on soldiers; it becomes a game of creation.”
We are no longer speculating about AI warfare. It has arrived. The “Silent Quds” operation shows us a future where algorithms decide targets, machines pull triggers, and wars are won in minutes rather than months.
This is a technological revolution—but also an ethical and geopolitical one. As AI takes deeper control of warfare, the world has entered a new, unpredictable chapter of human conflict.
Yesterday afternoon, everyone was slacking off and discussing that the U.S. military might follow the example of the Iraq War and launch a decapitation strike at night using its technological advantages.
Iran time is four and a half hours behind ours, so it should be morning here by then.
Although it was somewhat expected, I still didn’t think it would happen…
Early this morning, all media really flooded the screens with the news that Iran’s Supreme Leader Khamenei was confirmed dead after being attacked.
The death of Khamenei is a political event.
But the way that led to his death is a technological milestone.
The U.S. military’s decapitation operation targeting Khamenei, codenamed “Silent Holy City,” involved only one MQ-9B “Sea Guardian” drone, an 8-person special forces advance team. The core combat force was the AI combination of Claude 3 Opus large model and Palantir Foundry defense platform, as well as the JADC2 system integrated into the CENTCOM operational workflow.
The entire process took only 11 minutes and 23 seconds, achieving results with zero casualties, precise elimination of the target, and zero collateral damage.
The first time AI deeply intervened in a human war, it brought us such sci-fi-like data.
Brand-new structure
As early as early February, the U.S. military had already completed the deep integration of Claude and the Palantir system, and secretly connected it to the operational workflow of the U.S. Central Command, which is equivalent to equipping the war machine with a “super brain” plus a “neural center”.
Let’s first talk about the backgrounds of these two to prevent everyone from being confused.
Claude (Gov version). A large language model developed by Anthropic, and currently the only large model authorized by the Pentagon to operate on classified networks.
The Claude 4 Opus version deployed this time is a government-customized edition specifically optimized for military scenarios. It adopts a hybrid reasoning architecture, supporting dynamic switching between fast response mode and extended thinking mode. The context window has been expanded to 1M tokens, enabling it to process multi-modal inputs including text, images, and audio. Its AI security level reaches ASL-3, with enhanced protection specifically for high-risk military scenarios. It focuses on strengthening three core capabilities: unstructured data processing, multi-language semantic parsing, and combat scenario simulation.
The Claude Gov version achieves a real-time transcription accuracy of 98.7% for Persian calls, parses encrypted internal documents of the Iranian Revolutionary Guard 82 times faster than human analysts, and has an error rate of only 2.3% in simulating combat scenarios, far surpassing any AI analysis tool previously used by the U.S. military.
Palantir Gotham. The core product of Palantir, it received investment from the venture capital arm under the CIA as early as 2005. In 2025, its revenue exceeded 4.5 billion U.S. dollars, half of which came from U.S. military and government orders.
Its trump card lies in its unique dynamic ontology technology and full-link data integration capability, which can break down intelligence data silos, convert multi-source heterogeneous data such as satellite images, electronic signals, human intelligence, and social network data into intuitive entities of “person – location – action trajectory”, and then restore the complete activity trajectory of the target through knowledge graph and correlation analysis.
In this operation, Palantir also utilized the MetaConstellation platform, which can automatically dispatch hundreds of commercial and classified satellites to achieve 24/7 all-round monitoring. Even if the target is hiding underground, it can capture subtle traces of movement.
The U.S. military, through Palantir’s AIP, has deeply embedded the Claude large model into the Gotham platform, forming a seamless connection of “data collection – data analysis – decision generation – instruction issuance,” which is directly integrated into the operational workflow of the Central Command.
To put it simply, Claude is responsible for thinking and analysis, Palantir handles data integration and instruction execution, and the U.S. Central Command only takes charge of the final confirmation. The three form a closed loop.
It has completely broken the information barriers in traditional military operations where information is transmitted layer by layer among the “intelligence department, command department, and operational department”, and realized “real-time data sharing, real-time decision-making, and real-time issuance of orders”.
How terrifying would this brand-new combat structure be when applied to actual warfare?
AI + Military Industry Development Framework, Source: Guosheng Securities
Full kill chain closed loop
The operational chain of AI follows the principle of “data-driven, algorithm-led, with humans only making final authorization” at every step.
First step, intelligence perception.
The core of war is a contest of information gaps.
In this operation, Claude integrated more than 20 types of heterogeneous intelligence data, covering six dimensions: space-based, air-based, land-based, sea-based, cyberspace, and human intelligence. The total data volume reached 2.3 PB, involving 120 million intelligence fragments.
With this amount of data, if it were handed over to a traditional human intelligence team, assuming each person processes 1,000 pieces of intelligence per day, it would require 328 people working continuously for 100 days to complete. However, Claude finished full-scale cleaning, correlation modeling, and extraction of high-value information in just 90 minutes.
Palantir Foundry plays the role of a data infrastructure, processing 1.2 million pieces of heterogeneous data per second and supporting over 100,000 concurrent accesses. Meanwhile, it has built a relational graph of Iran’s top-level figures’ relationships, movement trajectories, and security system connections through knowledge graph technology, laying the foundation for subsequent target locking.
The second step: target locking.
Khamenei’s security system adopts anti-reconnaissance strategies of “irregular location changes, multi-level protection, and full-band electromagnetic shielding”, with the error radius of his daily activity range originally being more than 5 kilometers.
Meanwhile, on his way from his residence to the mosque for morning prayers, his vehicle passes through an air defense blind spot, and the gap between shifts of the security team is only 3 minutes.
That is to say, the assassination operation only has three minutes.
The above are the two core difficulties that make traditional decapitation operations difficult to achieve.
But tiny variables that are barely perceptible to humans appear as huge red lights in the eyes of AI.
For example, the average speed of Khamenei’s security convoy when passing through a certain intersection in the past three months was 45 km/h, but on the evening of February 28, the convoy came to a 1.2-second halt 3 kilometers away from the target point.
Only AI can detect anomalies in an extremely short time.
Claude built a time-series prediction model based on Khamenei’s movement trajectory, work and rest patterns, the shift schedule of his security team, and vehicle routes over the past 6 months. Incorporating variables such as Iranian holidays, religious activities, and weather conditions, it generated over 1,000 possible action paths, real-time excluded paths with a probability lower than 1%, and ultimately improved the prediction accuracy of the action window to 98.7%.
Palantir then superimposed Tehran’s urban geographic data, air defense system deployment data, and electromagnetic environment data with AI-predicted movement paths. Through high-precision physical simulation, it identified an air defense blind spot only 800 meters long.
Under the continuous modeling and trajectory prediction of AI, this error radius was reduced to 500 meters, ultimately locking in an absolute action window of 3 minutes.
The third step, decision deduction.
The core of traditional decision-making lies in humans. Commanders need at least several hours to organize their staff teams to conduct plan deduction, battle damage assessment, and risk analysis.
Claude, however, generated 15 complete raid plans within 8 minutes. Each plan covers six modules: drone penetration routes, bombing parameters, cover positions for special forces, timing of electronic warfare suppression, evacuation paths, and emergency response measures.
And with a simulation deduction efficiency of 30 seconds per plan, combined with Palantir’s battle damage assessment model, it conducts quantitative scoring on the battle damage rate, collateral damage rate, and penetration success rate of each plan.
The finally recommended plan scored 98.2 points (out of 100 points).
Its core strategy is: utilize the stealth capability of the MQ-9B drone to conduct ultra-low-altitude penetration in the air defense blind zone, launch 2 AGM-114 “Hellfire” missiles to precisely strike the target vehicle, while the EA-18G “Growler” electronic warfare aircraft conducts full-band electromagnetic suppression to block the signal transmission of Iran’s air defense system, and the advance team of special forces conducts vigilance around to prevent the target from escaping.
More importantly, because AI is directly connected to CENTCOM’s JADC2 system, all tactical parameters in the plan can be synchronized to all combat units of the army, navy and air force in real time, and the delay of cross-domain data flow is controlled within 200 milliseconds.
MQ-9B drone, Source: China Post Securities
This is the foundation of the fourth step: battlefield coordination.
The biggest pain point on the battlefield is the “military service information barrier”: the army uses tactical internet, the navy uses satellite communication systems, data formats are incompatible, command transmission requires manual translation, and delays usually range from several minutes to even dozens of minutes.
Claude, with its natural language interface capability, converts tactical commands in different formats into unified natural language. Having obtained Palantir’s IL6 classified environment certification, it not only enables the second-level issuance of commands but also ensures that all data flows within the classified environment, preventing information leakage.
Actual combat data shows that the cross-domain command delay of this operation is within 3 seconds, and the response synchronization rate of each combat unit reaches 100%: When CENTCOM issues the “execute operation” command, the command is synchronized to the MQ-9B drone, EA-18G electronic warfare aircraft, and special forces advance team within 0.5 seconds; after 1 second, the EA-18G initiates full-band electromagnetic suppression to block the signal transmission of Iran’s air defense system; after 2 seconds, the MQ-9B drone adjusts to the penetration route and descends to an altitude of 50 meters for ultra-low-altitude flight; after 3 seconds, the special forces advance team enters the alert position.
How can a large and bloated “old-style” army respond in just 3 seconds?
Step 5, strike execution.
The core of the strike execution phase lies in AI terminal correction and precise control of collateral damage.
In this operation, the two AGM-114 “Hellfire” missiles launched by the MQ-9B drone were both equipped with AI terminal guidance modules. These modules can recognize target vehicles in real-time during flight, correct the flight trajectory, and control the strike error within 1 meter.
Meanwhile, Claude accurately calculated the missile’s explosive yield, shockwave range, and fragment splash distance through the superposition of high-precision physical simulation and urban geographic data, controlling the scope of collateral damage within 10 meters.
For example, there are a large number of civilian buildings around the target. However, due to the AI’s precise control of the bombing timing and explosion height, only the target was destroyed in the end. The surrounding buildings and pedestrians were not damaged, achieving zero collateral damage.
This has, to a certain extent, reduced the international public opinion pressure on the U.S. military.
Step 6, damage assessment.
Damage assessment is a direct application of AI image recognition.
Claude used image recognition technology to identify that the target had been completely destroyed within 0.3 seconds, with no possibility of survival for the target personnel. At the same time, it confirmed that there were no civilian casualties in the surrounding area, and immediately sent an assessment report of “strike successful” to the CENTCOM commander.
The entire process is independently completed by AI, which is 1,000 times more efficient than traditional manual image analysis (which takes several minutes or even tens of minutes).
The above is the complete kill chain closed loop of this AI-led decapitation operation.
From target locking to final damage assessment, the entire process took only 11 minutes and 23 seconds.
At this moment, the core logic of modern warfare has been completely changed.
The Game of Creation
In this operation, the amount of intelligence data processed by AI reached 2.3 PB, involving 120 million pieces of intelligence fragments, which completely exceeded the limit of human processing capacity.
Moreover, with the passage of time, this order of magnitude is bound to grow exponentially.
This means that modern warfare has entered an era of “data overload”, where the human brain can no longer handle it and can only rely on AI.
Without AI, it is impossible to process massive amounts of intelligence data or eliminate information gaps; without AI, it is impossible to achieve the ultimate compression of the kill chain or gain a speed advantage; without AI, it is impossible to achieve timely coordination in cross-domain joint operations or form a combat synergy.
The decapitation operation targeting Khamenei this time is the first actual combat verification of this principle.
We can probably imagine what future intelligent warfare will look like.
First, algorithmic confrontation.
Whoever has better algorithms, stronger data processing capabilities, and faster decision-making speed will be able to take the initiative.
Weapons of lethality such as fighter jets and missiles are just tools for executing algorithms. The strength of their combat effectiveness depends more on the quality of the AI algorithms they are equipped with.
The global military AI development data also confirms this trend.
In 2025, the global military AI market size reached 11.6 billion U.S. dollars.
From 2024 to 2031, the compound annual growth rate is expected to reach 14.49%, and the market size will exceed 35.54 billion US dollars by 2031.
Currently, the U.S. military is developing an “autonomous decision-making algorithm” to enable AI to independently make combat decisions; China is developing a “swarm coordination algorithm” to realize the swarm operations of unmanned aerial vehicles and unmanned boats; Russia is developing an “electronic countermeasure algorithm” to interfere with and suppress the enemy’s AI systems.
This is a contest among algorithm engineers.
Second, unmanned warfare.
In the current model, at the strike level, AI still requires human soldiers to execute.
But this situation may change soon.
In 2025, the global military drone market size exceeded 30 billion U.S. dollars; the U.S. military is advancing the CCA (Collaborative Combat Aircraft) program, applying for 789 million U.S. dollars in the fiscal year 2025, and planning to invest 28 billion U.S. dollars by 2029, equipping each F-35 fighter jet with 2-3 unmanned wingmen.
By 2030, unmanned combat platforms will account for 60% of the U.S. military’s platforms. The number of unmanned aerial vehicles in the Air Force will exceed that of manned aircraft, and the Navy’s unmanned boats and unmanned underwater vehicles will become core components of the aircraft carrier strike groups.
2025 Military AI-related Policies, Source: Toubao Research Institute
Third, low-cost warfare.
Traditional wars are typically “high-cost and high-casualty”. A local war can cost hundreds of billions of dollars and cause a large number of casualties.
And in this decapitation operation targeting Khamenei, the cost for the U.S. military was only 10 million U.S. dollars, which is 1/5 of that of traditional decapitation operations, and zero casualties were achieved.
The reduction in war costs and casualty rates will significantly increase the tendency of major powers to launch unilateral military operations.
For many years in the past, when a country waged a war, it had to consider the costs and casualties, and domestic anti-war sentiment would also restrict it.
In future intelligent warfare, military strikes against other countries can be launched at extremely low costs and with zero casualties. This will significantly reduce domestic anti-war sentiment, leading to a substantial increase in global military conflicts and a more tense geopolitical situation.
When AI is reduced to a tool of war and algorithms determine the direction of the world.
What will the future look like?
The first thing that comes to mind is the prophecy in the 56th hexagram of Tui Bei Tu:
Those that fly are not birds; those that dive are not fish. Battles do not depend on soldiers; it is all a game of creation.
Disclaimer: The views in this article are based on public reports and analysis as of March 2026, for discussion and reflection only, not investment or military advice.
