A handgun, two riders, and a congested street outside a mosque in Rawalpindi or Karachi: that combination has done the work of an entire counter-terrorism doctrine for the better part of a decade, and it has done that work because it is cheap, deniable, and human. The question that now sits in front of every intelligence service watching this campaign is whether the next decade keeps the human in the picture at all.

Four technologies are converging on the practice of clandestine elimination, and each of them attacks a different part of the method that India has used to hunt wanted men on Pakistani soil. Algorithmic target identification is collapsing the months of patient watching that used to precede a strike. Small uncrewed aircraft are removing the need for a shooter to ever stand within pistol range of a victim. Cyber sabotage is opening the possibility of causing death through infrastructure rather than through bullets. Biometric tracking is making the disguises, false identities, and quiet relocations that protected hunted men for years close to useless. By 2030 the motorcycle-borne shooting may look less like the cutting edge of statecraft and more like the last generation of a craft that machines are about to inherit.

This article forecasts that transformation, technology by technology, and tests it against a sober objection. Paul Scharre, the former Army Ranger who drafted the Pentagon’s first directive on autonomy in weapons and later wrote the field-defining book on the subject, has spent years warning that the gap between what these systems can do in a laboratory and what they can do reliably in the chaos of a real operation is wide and slow to close. James Johnson, who studies how machine intelligence interacts with nuclear stability, argues that the danger is less the capability itself than the temptation it creates to act faster and with less reflection than humans should. Both cautions belong in any honest projection. What follows treats the four technologies as trends with momentum behind them, not as prophecies, and it keeps returning to the same governing question: when the surveillance, the approach, and the trigger pull can all be handled by software and machines, what survives of the signature method that built India’s reputation as the most disciplined practitioner of quiet killing in the modern era?

The Four Technologies Reshaping the Hunt

The campaign that InsightCrunch has documented across dozens of cases rests on a sequence that has barely changed since the first eliminations of 2022. A wanted man is located. His routine is established through weeks of observation. His prayer schedule, his route to the bazaar, the house where he sleeps, and the company he keeps are all mapped. A two-man team on a motorcycle closes the final distance, fires, and disappears into traffic before the body has settled. The unknown gunmen pattern is, in the end, a pattern precisely because each stage of it is performed the same disciplined way every time.

Each of the four emerging technologies dissolves one stage of that sequence. Artificial intelligence attacks the location and observation phase, because pattern-recognition software can sift through communications metadata, satellite imagery, financial records, and social media at a speed and scale that no team of human analysts can match. Autonomous micro-drones attack the approach and the trigger, because a machine the size of a bird can carry a small charge to a window without a single operative leaving home. Cyber operations attack the very definition of a strike, because a manipulated insulin pump, a sabotaged vehicle control system, or a poisoned industrial process can kill without anyone ever pointing a weapon. Biometric tracking attacks the foundation underneath all of it, because a man whose face, gait, iris, and voice are all enrolled in a searchable database cannot hide behind a new name and a relocated household the way Zahoor Mistry hid for two decades before the campaign found him.

Scharre’s research is the right anchor for understanding why this is happening now rather than later. By his count, at least thirty nations already field weapons capable of searching for and striking targets on their own, most of them defensive systems that operate under human supervision. The threshold that matters is not whether a machine can pull a trigger, since machines have done that since the homing torpedoes of the Second World War. The threshold is whether a machine can decide who the trigger is pulled against. That decision, the identification of a specific human being as a legitimate object of lethal force, is the part of covert work that has always demanded the most patience, the most skill, and the most deniable human judgment. It is also the part that the next decade of technology is most aggressively trying to automate.

The convergence is not accidental, and understanding why it is happening now clarifies how fast it will move. Each of the four technologies was developed for a purpose far larger than covert killing. Algorithmic targeting grew out of the need to process surveillance data that had outgrown human analysts. Small drones grew out of consumer robotics and the demands of conventional war. Cyber capability grew out of decades of investment in network warfare. Biometric identification grew out of border control, policing, and the commercial security industry. None of these fields was built to serve assassination. All of them, having been built, can be turned to it, and that is the pattern that makes the forecast credible: covert services do not have to fund the basic research, because the basic research is being funded for them by far larger markets and far larger military programs. They only have to adapt mature tools to a narrow purpose.

That adaptation is also cheaper and faster than building a capability from nothing. A service that wants an algorithmic targeting model does not start from an empty page; it starts from the machine-learning techniques, the hardware, and the trained personnel that a global technology industry has already produced. A service that wants flying machines can buy loitering munitions on an export market or copy designs that have been thoroughly documented in the wars of the 2020s. The barrier to entry for technological covert killing is falling year by year, and a falling barrier means the question is not whether the major intelligence services adopt these methods but how quickly, and how many smaller actors follow them.

The historical frame worth keeping in view is that covert killing has always tracked the available technology. The poisons of the early Cold War, the rifle and the car bomb of its middle decades, the precision of the modern era: each generation of clandestine work used the tools its era produced. India’s motorcycle method is the current generation’s answer, and it is an unusually elegant one, because it solved the problem of deniability with nothing more advanced than traffic. The four technologies described here are simply the next set of tools, and the only genuinely novel feature of this generation is the depth of the question it raises. Earlier tools changed how a person killed. These tools raise the possibility of removing the person from the killing altogether, and that is a difference of kind, not merely of degree.

There is a reason intelligence services are pursuing this rather than resisting it. Every stage of the traditional method carries a vulnerability. Human surveillance teams can be spotted, followed, and rolled up. Shooters can be captured alive, and a captured shooter is a confession waiting to happen. Each operative who knows the plan is a leak that has not happened yet. A campaign that depends on dozens of trained people moving through hostile cities is a campaign that bleeds secrecy at every joint. The promise of the four technologies is not merely speed. It is the removal of the human witness, and the removal of the human witness is the oldest dream of clandestine statecraft. What India built with motorcycles, others are trying to build with code, and the contest between those two visions of the shadow war will define the field through 2030 and beyond.

Artificial Intelligence and the End of Human Surveillance

The single most labor-intensive element of every elimination documented in this series is the surveillance that precedes it. To shoot Abu Qasim outside the Al-Qudus mosque in Rawalakot at the moment he arrived for prayers, a team first had to learn that he prayed there, which prayer he attended, how reliably he kept the schedule, and which approach offered both a clean shot and an escape into traffic. That knowledge cost at least two weeks of patient watching by people who themselves risked exposure every hour they spent in position. Artificial intelligence is built to make that two weeks unnecessary.

The clearest preview of what algorithmic targeting looks like in practice comes from Israel, where the military deployed two machine systems during its Gaza campaign that have since been examined in detail by journalists and legal scholars. The first, known as the Gospel, was designed to mark buildings and structures that intelligence analysts could then assess as potential objectives. The second, called Lavender, did something far more consequential: it marked people. Whereas the Gospel marks buildings and structures that the army claims militants operate from, Lavender marks people and puts them on a kill list. Both systems were built by Unit 8200, the Israeli signals intelligence corps, and both fused enormous volumes of surveillance data into outputs that human officers were expected to act on quickly. According to the reporting that first exposed the program, Lavender initially identified around 37,000 Palestinian men linked to Hamas or Palestinian Islamic Jihad.

The numbers attached to Lavender’s use are the part worth dwelling on, because they describe a future that India’s planners are certainly studying. In practice, soldiers often made rapid decisions, sometimes taking as little as 20 seconds, based on the system’s output to determine whether to target an identified individual. That same reporting found that the program carried an error margin of up to 10 percent, and soldiers frequently followed its output without question. Strip away the specific conflict and what remains is a template: a machine converts raw intercepted data into a ranked list of human beings, and the human role shrinks from investigator to reviewer, from the person who builds the case to the person who spends twenty seconds confirming a gender before authorizing a death.

The American equivalent shows the same trajectory from a different starting point. Project Maven began in 2017 as a relatively modest effort to use machine vision to sort the flood of drone footage that human analysts could not process fast enough. It did not stay modest. Project Maven has evolved from being a sensor data analysis program in 2017 to a full-blown AI-enabled target recommendation system built for speed, and its operators can now sign off on as many as 80 targets in an hour of work, versus 30 without it. The phrase the United States military uses for what these systems compress is the “sensor-to-shooter timeline,” and the entire institutional logic of military AI is bent toward making that timeline shorter. Systems like Lavender are the manifestation of a broader trend that has been underway for a good decade, and elite units are far from the only ones seeking to implement more AI-targeting systems into their processes.

Project Lavender, Gospel, and Maven were built for high-intensity conventional war, where the targets number in the tens of thousands and speed is the dominant value. A campaign of selective elimination is the opposite problem. The shadow war does not need to identify 37,000 men. It needs to find one man, confirm beyond reasonable doubt that he is the man, and locate him precisely enough for a strike. That difference does not make AI less relevant to covert work. It makes it relevant in a sharper way. A targeting model tuned not for volume but for certainty could take the months of human surveillance behind a single elimination and replace them with continuous algorithmic monitoring of a known individual across every digital surface he touches. His phone’s metadata, the facial-recognition hits from cameras he passes, his pattern of purchases, the movement of the people around him, and the rhythm of his communications would all feed a model whose job is to answer one question: where will this specific man be, predictably, alone enough, and long enough, for a strike to succeed?

The data feeding such a model is already being generated, which is the detail that makes the 2030 projection credible rather than fanciful. A wanted man in a Pakistani city leaves a trail he cannot fully control. His mobile phone reports its position to cell towers whether or not he makes a call. His associates carry phones of their own, and the pattern of which devices cluster near his establishes his network without a word being intercepted. Money moves through accounts and informal transfer systems that leave records. Vehicles he rides in pass cameras. Family members post to social media. None of these traces is decisive on its own. An algorithm built to fuse them is decisive in combination, because the value of machine analysis is precisely the ability to find the signal that no single human analyst, looking at any single stream, would ever assemble. The surveillance that once required a team of watchers risking exposure on a Pakistani street becomes, in this model, a continuous background process running on a server in another country.

What this does to the structure of an intelligence service is as consequential as what it does to the speed of an operation. The classical hunt for a wanted man was a project: a case officer was assigned, a team was built, resources were committed, and the work had a beginning and an end. Algorithmic monitoring has no such shape. It is a standing capability that watches everyone of interest all the time, and an elimination becomes not the culmination of a dedicated project but the moment a supervisor decides that a continuously maintained picture is good enough to act on. That is a profound change in tempo. It compresses the distance between deciding that a man should die and being able to kill him from months to, potentially, hours, and the doctrine that governs India’s covert work was not written for a tempo that fast.

India’s particular advantage and particular difficulty both come from the same source. The advantage is a software industry of genuine global weight and a vast pool of engineers fluent in exactly the machine-learning techniques an algorithmic targeting system would require. The difficulty is that an intelligence service does not become a software company by deciding to. Building a targeting model that a service can trust with lethal decisions requires not only engineers but data of high quality, validation that is rigorous rather than convenient, and a culture willing to treat the model’s outputs with disciplined skepticism rather than the deference that the Lavender reporting found among soldiers under pressure. The countries that gain the most from algorithmic targeting will be the ones that pair technical depth with institutional caution, and institutional caution is the rarer commodity.

By 2030 the realistic projection is that the surveillance phase of a covert elimination shrinks from weeks of human exposure to a query against a system that has been watching continuously. The intelligence officer stops being the analyst who assembles the picture and becomes the supervisor who decides whether to trust it. That shift is precisely what worries Scharre, whose central argument is that machines can be made to perform narrow tasks brilliantly while remaining incapable of the contextual judgment that tells a human when the brilliant answer is wrong. A model that has correctly tracked a wanted man for six months can still misread the day he lends his phone to his brother, and a system optimized to compress the sensor-to-shooter timeline is, by construction, a system that gives the human reviewer less time to catch that error. The agency that adopts algorithmic targeting buys speed and scale. It pays in the erosion of the deliberate, slow, deniable human judgment that has so far kept India’s campaign as clean as its doctrine required it to be.

Autonomous Micro-Drones and the Vanishing Assassin

If artificial intelligence dissolves the surveillance phase, autonomous flying machines dissolve the most dangerous phase of all, which is the moment a human being has to physically close the distance to a target and pull a trigger within arm’s reach. That moment is where shooters are caught, where witnesses see faces, and where a clean operation turns into a diplomatic catastrophe. A machine that can carry a charge to a target on its own removes the moment entirely.

The technology is not speculative. It is already in service, already exported, and already battle-tested. Turkey’s STM Kargu, a portable rotary-wing attack drone, entered service with the Turkish armed forces in 2018 and has since been sold across four continents. A United Nations panel examining the Libyan civil war reported that a Kargu-2 had been used there in a way that should be read carefully by anyone forecasting covert work: the panel described the machine being used to attack targets without requiring data connectivity between the operator and the munition, in effect a true fire, forget and find capability. That single sentence, buried in a Security Council document, may describe the first time in history that a machine selected and attacked a human being with no human confirming the specific kill. The manufacturer states the system operates under a man-in-the-loop principle, but the Libya finding shows how thin the line between supervised and autonomous already is.

The size trend is the one that should most concern a protective service. The drones that defined the war in Ukraine were, for the most part, large enough to see and hear. The trajectory of development runs relentlessly toward the small. Micro-drones the size of a sparrow, carrying a charge sufficient to kill one person at close range, are an active area of work in several defense industries, and the physics that makes them possible is the same physics that has shrunk cameras, batteries, and processors in consumer electronics for two decades. A machine that small does not announce itself. It does not need an airstrip or a launcher larger than a backpack. It can be carried into position in an ordinary vehicle, released kilometres from a victim, and recovered or expended without anyone noticing it was ever airborne. The motorcycle was chosen for India’s campaign partly because it is unremarkable in a Pakistani street. A micro-drone is more unremarkable still, because it need not be in the street at all.

There is also the matter of what happens after the strike. A two-man team that completes a shooting must survive the most dangerous minutes of the work, the escape through traffic with a description of them already circulating. A flying machine has no such problem, and this asymmetry reshapes the risk calculation of an entire campaign. The planners who built the unknown gunmen pattern had to weigh every operation against the possibility that a shooter would be caught and that a caught shooter would talk. That single fear has shaped the geography, the timing, and the tempo of the human campaign. Remove it, and the campaign can reach figures it currently leaves alone: men in better-guarded locations, men in cities where a motorcycle escape is impractical, men whose elimination by human hands would carry an unacceptable risk of capture. The machine does not merely make existing operations safer. It expands the set of operations a service is willing to attempt.

The economics deserve attention because they cut against the intuition that advanced technology is expensive. A purpose-built military aircraft can cost a great deal, but the loitering munitions that have proven decisive in recent wars are often cheap, and the commercial quadcopters that armies have converted into weapons cost less than a motorcycle. Low cost has a strategic consequence. A method that is cheap enough can be used at scale, and a method that can be used at scale changes a campaign from a series of carefully husbanded operations into something closer to a routine. The discipline of the human campaign came partly from scarcity: trained teams were a limited resource, and a limited resource is spent carefully. A cheap, abundant, expendable machine removes that natural brake, and the removal of natural brakes is a theme that runs through every one of these technologies.

The war in Ukraine has accelerated everything. The conflict turned the loitering munition, a drone that orbits an area until it finds something worth destroying and then dives into it, from a niche weapon into the defining tool of the battlefield. American Switchblade munitions small enough to fit in a backpack were shipped by the hundreds. Russian and Ukrainian forces converted commercial racing quadcopters into precision weapons by strapping explosives to them. The most significant development for the future of covert work is what happened next: the machines started navigating and choosing on their own. Ukrainian military intelligence documented a Russian loitering munition, designated the V2U, that can autonomously search for and select targets using artificial intelligence, with its processing built on a Chinese-made carrier board and AI module. The same machine was reported to have a terrain-matching capability for navigation in jammed environments, using computer vision to compare its camera footage against pre-loaded imagery. A drone that can find its way without satellite navigation and pick its own target without a radio link is a drone that an enemy cannot jam, cannot trace back to an operator, and cannot stop by capturing the person who launched it.

Swarming compounds the threat. STM now advertises that its Kargu units can operate in coordinated groups built on what the company calls swarm-intelligence software, with the machines performing real-time object detection and classification, dividing into sub-groups, and sharing target priorities among themselves. Ukrainian observers have described groups of Russian V2U munitions forming up before a dive, using markings on their wings to hold formation. The same firm that makes the Kargu has reportedly worked on facial-recognition payloads, which points directly at the covert application: not a swarm sweeping a battlefield, but a single small machine carrying a face in its memory, loitering near a known location, and acting the moment the right face appears.

For a campaign like India’s, the autonomous micro-drone is the natural successor to the motorcycle, and the drone exchanges seen during the 2025 conflict already proved that both sides treat uncrewed systems as core capability rather than novelty. A bird-sized machine launched from a vehicle two kilometres away, navigating without a traceable signal, identifying a target by face, and delivering a charge the size of a grenade solves nearly every problem the two-man team carries. There is no shooter to capture. There is no face for a witness to describe. There is no escape route to plan, because the launch point and the strike point are never the same and the operator is never near either. The deniability that India currently buys with traffic and chaos could be bought far more cleanly with a machine that, if it fails, reveals only itself.

The honest forecast has to include the limits, and this is where Scharre’s caution earns its place. Small drones remain fragile, weather-dependent, and modest in payload. Facial recognition from a moving aerial platform against a target who may be looking down, wearing a shawl, or standing in a crowd is far harder than recognition from a fixed border camera. A swarm is also a coordination problem, and coordination problems fail in ways their designers did not anticipate. The motorcycle method will not vanish in a single year because the drone is not yet reliable enough to fully replace it. The trajectory is unmistakable nonetheless. Every year the machines get smaller, cheaper, smarter, and less dependent on the radio links and satellite signals that make them vulnerable, and every year the comparison between drone and human methods shifts further toward the machine.

Cyber Operations and Killing Through Infrastructure

The third technology changes not the speed or the safety of a strike but the very meaning of the word. A cyber operation does not put a weapon near a victim. It reaches into the systems the victim depends on and turns one of them into the cause of death. The bullet, in this model, is replaced by a line of code, and the question of who fired becomes nearly unanswerable.

The proof that code can cross into the physical world and destroy machinery is more than fifteen years old. Stuxnet, the malicious software discovered in 2010 and widely attributed to a joint American and Israeli program, was built to attack the centrifuges enriching uranium at Iran’s Natanz facility. It did not merely steal data or disable computers. It quietly altered the speed of the centrifuges while reporting normal readings to the engineers watching the controls, causing the machines to tear themselves apart over months in a way that looked, from the inside, like ordinary mechanical failure. Stuxnet established two principles that the future of covert work is built on. The first is that software can produce physical destruction. The second, and the more important for clandestine purposes, is that it can do so while disguising the destruction as an accident.

Extending that logic from a centrifuge to a human being is not a large conceptual leap, and the surface area is growing every year. A modern car is a network of computers governing its brakes, steering, and acceleration. A hospital patient is increasingly surrounded by networked devices, from infusion pumps that meter drugs to implanted cardiac units that can be reprogrammed wirelessly. An apartment is filling with connected systems. Each of these is, from the perspective of a cyber operation, a potential instrument. A vehicle made to lose its brakes on a mountain road, an insulin pump made to deliver a fatal dose, a building’s gas system made to fail at night: each would read, to investigators, as misfortune rather than murder. The phrase cyber assassination sounds like fiction, but the components it would require are ordinary consumer technology with known vulnerabilities, and the intelligence services capable of writing software like Stuxnet are equally capable of writing something smaller and more personal.

History already offers grim previews of how a digital intrusion can become a physical death, even where killing was not the original intent. The sabotage of industrial control systems, the manipulation of power grids that left civilians without heat in winter, and the documented vulnerability of medical devices to wireless reprogramming all demonstrate that the boundary between the digital and the physical is thoroughly porous. Security researchers have for years shown, in controlled demonstrations, that connected cars can be made to brake or accelerate against the driver’s input and that implanted cardiac devices can be reached and altered without physical contact. These demonstrations were warnings, published to force manufacturers to repair flaws. An intelligence service reads the same research as a catalogue.

The strategic logic of cyber killing is best understood by comparing it against the other three technologies on the single dimension that matters most to a covert service, which is what the method leaves behind. An algorithmic targeting system leaves a decision trail. An autonomous drone leaves wreckage. A biometric hunt leaves a body with an obvious cause of death. A well-executed cyber operation leaves a death that is never investigated as a killing at all, because no one classifies it as one. For a campaign that has spent years issuing denials about killings that were transparently killings, as the pattern documented across India’s shadow war shows, the appeal of a method that removes the need for a denial is not marginal. It is the closest the trade has ever come to its founding fantasy: an outcome achieved with no evidence that anyone achieved it.

The defensive side of cyber operations belongs in any honest account, because it points to the limit of the method. Connected systems can be hardened. A protective service that understands the threat can keep its high-value figures away from networked vehicles, networked medical devices, and networked homes, reverting them to a deliberately analog existence. This is not a hypothetical countermeasure; it is the obvious one, and it costs almost nothing. The result is that cyber killing, more than any of the other three technologies, can be defeated by a target who simply chooses to live without the technology that would be used against him. That makes it a powerful tool against the modern and the connected and a weak tool against the disciplined and the primitive, and the men at the top of India’s target list have generally been disciplined.

What makes this attractive for covert killing is not lethality, because a pistol is perfectly lethal. It is attribution. Every other method in this article leaves a physical trace. A drone leaves wreckage. A shooting leaves a casing, a body, a cause of death that is obviously violent. A cyber operation, done well, leaves a death that no one even classifies as a killing. There is no crater, no projectile, no enemy operative to hunt. There is a grieving family, an unremarkable accident report, and an intelligence service that has achieved the oldest goal of the trade, which is an outcome that cannot be proven to be an outcome at all. For a campaign that has spent years denying involvement in killings that were plainly killings, the appeal of a method that denies the killing ever happened is considerable.

The constraints here are real and they cut in a specific direction. A cyber operation against a single individual requires that the individual use vulnerable connected technology, and many of the men India has hunted live deliberately low-technology lives in Pakistani towns where a networked car or a smart medical device is uncommon. A wanted commander who travels by ordinary motorbike and is treated at a basic clinic presents almost no cyber surface at all. This is the technology whose covert application is most limited by the profile of the typical target, and it is therefore the one least likely to displace the motorcycle for the hardened, low-technology figure. Its real growth lies elsewhere, against wealthier, more connected, more modern targets, and against the infrastructure that protects them. The trend that matters is the one Scharre identified years ago when he listed autonomous cyber weapons alongside robotic ships and drone armies as a single emerging category: software is becoming a weapon that acts in the physical world, and the services that master it gain a way to kill that the rest of the world cannot see.

Biometric Tracking and the Death of Anonymity

The first three technologies change how a strike is delivered. The fourth changes whether a target can ever be lost in the first place, and in the long run it may be the most decisive of all, because it attacks the single resource that every hunted man has always depended on, which is the ability to disappear.

For decades the standard survival strategy of a wanted militant was identity laundering. A new name, a forged identity document, a relocated household, a beard grown or removed, a quiet town far from where the man was known: these were enough to buy years, sometimes decades. Zahoor Mistry, eliminated in Karachi more than twenty years after the IC-814 hijacking, survived that long behind exactly such a constructed identity. Biometric tracking is the technology that ends this strategy, because the traits it reads cannot be changed by changing a name. A face has a measurable geometry. An iris has a unique pattern. A voice has a signature in its frequencies. And a body has a gait, a particular rhythm of walking, that belongs to one person.

The state of the technology in the middle of the 2020s already makes anonymity fragile. Border agencies have built comprehensive biometric collection into ordinary travel. United States Customs and Border Protection runs a Biometric Entry-Exit program that captures facial images, fingerprints, and iris scans from international travelers at airports and border crossings, integrating facial recognition, fingerprint matching, and iris scanning with AI-powered risk assessment. The databases behind these systems have expanded well past faces and fingerprints. Databases such as the Homeland Advanced Recognition Technology platform now store not only fingerprints and facial images but also iris and gait patterns. The same reporting documented mobile applications that let officers photograph a subject’s face or fingerprint and trigger a near-instant biometric match against multiple data sources. A capability that recently lived only at fixed checkpoints is moving into handheld devices that work anywhere their operators stand.

Gait recognition is the development that should most alarm anyone whose survival depends on staying unidentified, because it works at exactly the distances covert surveillance operates at and it resists exactly the disguises that used to work. A man can cover his face. He cannot easily change the way he walks. Researchers describe gait analysis as a biometric technique that permits recognizing an individual from a long distance, focusing on features such as movement, time, and clothing, and is highly useful in video surveillance scenarios. The frontier is the fusion of several traits into one identification, designed to work in the worst conditions. A 2025 research system built for the United States biometric research community combined face, body shape, and gait recognition for unconstrained biometric identification at long distances and elevated viewpoints, for applications including law enforcement, border security, and wide-area surveillance. Long distances and elevated viewpoints describe a drone. A wide-area surveillance system that can identify a man by the combination of his face, his build, and his walk, from the air, in poor light, while he believes he is anonymous in a crowd, is the precise instrument that turns a city into a place where a hunted man can be found.

China’s domestic surveillance build-out is the working model of what a biometric-saturated environment looks like, and it is the reference every intelligence service now studies. Hundreds of millions of cameras, many tied to facial-recognition systems, combined with identity registration that links a face to a name, an address, and a movement history, have produced an environment in which anonymity is not difficult but structurally impossible. The relevance to covert work is not that any single state will replicate the Chinese system wholesale. It is that the technical components of that system, the cameras, the recognition algorithms, the databases, the linkage of biometric traits to identities, are commercially available and spreading. A wanted man does not need to live under a comprehensive surveillance state to be caught by one. He only needs to cross the territory of any state that has installed the components, and the export of those components means the territory that is safe for a hunted man shrinks every year.

Voice adds a dimension that disguise cannot easily defeat. A man can alter his appearance and consciously change his walk for short periods, though not for long. He cannot easily change the acoustic signature of his own voice, and intelligence services have collected voice samples through intercepted calls for as long as telephones have existed. A voiceprint matched against an intercepted call places a target on a particular line at a particular moment, and the fusion of a voiceprint with location data from the same call narrows a search from a country to a neighbourhood. The biometric net is not one technology. It is the convergence of face, iris, gait, and voice into a single mesh, and the defining property of a mesh is that slipping past one strand still leaves a man caught in the others.

What biometric tracking ultimately attacks is time. The protective infrastructure that has sheltered wanted men, the new names and the relocations, worked because it bought years, and years were enough for the political weather to change, for a campaign to lose momentum, for a hunted man to die of natural causes in his bed. Biometric enrollment collapses that purchased time. Once a man’s face, iris, gait, and voice sit in a searchable system, the protection a false identity offers is measured not in years but in the interval between sensor encounters, and in a world filling with sensors that interval shrinks toward zero. The deepest change biometric tracking brings to the shadow war is not that it makes any single elimination easier. It is that it removes the possibility of waiting a hunt out, and removing that possibility changes the calculation of every man who has ever believed that if he stayed hidden long enough, the danger would pass.

For the future of the shadow war the implication is stark. The protective infrastructure that Pakistan has provided to wanted men, the new identities and the relocations and the quiet towns, is built to defeat human investigators who must be told a name before they can begin. Biometric tracking does not need the name. It needs the body, and the body is enrolled the first time the man crosses a monitored border, appears on a monitored street, or is captured by a monitored camera. Once a target’s biometric signature exists in a searchable system, the burden of the hunt inverts. The campaign no longer has to find where a man is hiding. It has to wait for the man to walk past a sensor, and in a world filling with sensors that wait is short. This is also the technology with the gravest implications beyond the battlefield, because the same systems that strip anonymity from a wanted terrorist strip it from journalists, dissidents, and ordinary citizens, which is why the debate over biometric surveillance has become one of the central civil-liberties arguments of the decade. The covert operator and the protester are made visible by the identical machine.

How the Four Technologies Rewrite the Covert Playbook

Examined separately, each technology improves one stage of an operation. Examined together, they do something more radical: they describe a covert elimination in which a human being need not be present at any stage, and that combined effect is what genuinely rewrites the playbook rather than merely upgrading it.

Trace a single hypothetical operation through all four. A biometric system, fed by border crossings and city cameras, registers that a wanted man has surfaced in a particular town and confirms his identity by the fusion of his face and his gait. An artificial-intelligence model that has been monitoring his digital and physical pattern establishes, without a single human watcher on the ground, that he visits a specific location at a predictable hour. An autonomous micro-drone, carrying that man’s biometric signature in its memory, is launched from a vehicle well outside the town, navigates without a traceable signal, loiters, recognizes the face, and strikes. Or, if the man is wealthy and connected enough to present the surface, a cyber operation reaches the vehicle that carries him and the strike registers as an accident. At no point in either version does an operative enter the town. At no point is there a shooter to capture, a watcher to spot, or a face for a witness to describe. The campaign’s exposure has been reduced from a dozen human beings to a piece of machinery and a few lines of code.

That is the doctrinal transformation, and its core is the removal of the human witness from the act itself. Consider what each technology eliminates from the current method. The surveillance team, the most exposed and longest-deployed human element, is replaced by an algorithm. The shooter, the element whose capture is most catastrophic, is replaced by a machine or by software. The escape, the phase that requires planning and luck, becomes irrelevant because there is nothing at the scene that needs to escape. The witness, the uncontrollable variable, sees a drone or sees nothing at all. Every category of human vulnerability that a counter-intelligence service like Pakistan’s ISI could exploit, the field officer who can be followed, the shooter who can be broken in interrogation, the witness who can describe a face, is engineered out of the operation.

The shift also changes who conducts covert work and what skills the trade rewards. The classical intelligence officer was a runner of human agents, a reader of human behavior, a person whose value lay in tradecraft built over a career. The covert service of 2030 needs data scientists who can build and tune targeting models, drone engineers, cyber specialists, and biometric analysts. The center of gravity moves from the field officer to the technician, and from the safe house to the server room. This is a change that institutions find genuinely difficult, because it asks an intelligence culture built around human sources and human judgment to hand its most consequential decisions to people who never meet an agent and to systems that never explain themselves. The agency that manages this transition well gains a covert capability of unprecedented reach. The agency that manages it badly automates its own mistakes at machine speed.

The transformation also redraws the geography of covert work. The human campaign is constrained by where its people can safely go. A surveillance team must be able to live in a city without standing out, a shooter must be able to reach and leave a location, and a support network must exist to move people and weapons. Those constraints have kept the documented eliminations concentrated in particular kinds of places: cities with enough chaos to absorb a strike, towns close enough to a border to permit an exit. A technological campaign is constrained by entirely different things. An algorithm does not need to be physically present anywhere. A drone needs only a launch point within its range. A cyber operation needs only a network connection. The map of where a service can reach expands dramatically, and the deep interior of a country, the garrison towns and protected compounds that a human team could never safely penetrate, stops being a sanctuary.

Precision about what does not change matters here, because a transformation narrative tends to oversell itself. The decision to kill a particular person remains a human decision, or it should. The intelligence that establishes a man’s identity and his guilt still rests, at its foundation, on judgments about evidence that no current system makes on its own. The political authority to run a covert campaign, the strategic choice of who belongs on a target list, and the willingness to accept the consequences of exposure are all human and institutional matters that no drone or algorithm touches. The technologies automate the execution of covert killing. They do not automate the responsibility for it, and a service that confuses the two, that allows the ease of execution to erode the seriousness of the decision, has not modernized its tradecraft. It has degraded it.

There is a quieter institutional consequence that follows from all of this, and it concerns secrecy itself. A human covert campaign is secret in a particular way: a small number of people know, and the secret holds as long as those people hold it. A technological campaign is secret in a different and more fragile way, because it depends on systems, and systems generate records, logs, and artifacts whether or not anyone intends them to. A targeting model has training data. A drone has a manufacturing trail. A cyber tool has code that can be captured and studied. The paradox is that automating a campaign to remove the human witness may simply trade a human point of failure for a technical one, and technical points of failure, once discovered, tend to reveal not a single operation but the entire system behind it. A captured shooter exposes one job. A captured targeting model could expose a doctrine.

There is a counter-current worth naming, because the playbook does not rewrite itself in only one direction. The same technologies that empower the hunter also empower the hunted and the defender. Counter-drone systems, signal jammers, and the physical netting that Ukrainian forces strung over roads are all responses that work against machines. Biometric systems can be fed false data. Targeting models can be deceived by adversaries who understand how they reason. A protective service that adapts can make the technological campaign almost as difficult as the human one. The future is not a straight line from the motorcycle to the autonomous machine. It is a contest, and the intelligence war between ISI and RAW will be fought in this technological domain as fiercely as it has ever been fought in the human one.

The Ethics of Algorithmic Killing

A forecast that described only capability would be incomplete and, worse, misleading, because the most important question raised by these four technologies is not whether they will work. It is whether a state should want them to, and the answer is genuinely contested among the people who have thought hardest about it.

One side of the argument holds that the technologies make covert killing more precise, more proportionate, and therefore more humane. A machine does not panic, does not take revenge, and does not misfire under stress. An algorithm that has tracked a target for months may identify him with more confidence than a human team working under pressure in a hostile city. A drone carrying a small charge to a specific window may cause less collateral harm than a firefight in a crowded bazaar. By this reading, automation is a moral improvement, and a state that can kill its designated enemies with fewer bystander deaths and fewer errors has an obligation to use the better tool. This is close to the position Scharre himself stakes out, with an important qualification: he argues that technology should be embraced where it makes war more precise and more humane, but that when the choice is genuinely one of life and death there is no adequate replacement for human judgment. Precision is welcome. The surrender of the final decision is not.

The other side of the argument is the one that should give planners the most pause, and it is the one James Johnson has pressed hardest. The danger of these technologies is not that they kill the wrong people, though they will sometimes do that. The danger is that they make killing easy. Every element of friction in the current method, the cost of training a team, the risk to that team, the slow accumulation of human judgment, the political weight of ordering men into danger, acts as a brake. Each of those brakes forces a decision-maker to want the killing badly enough to pay a real price for it. Remove the team, remove the risk, remove the friction, and the price collapses. A capability that lets a state eliminate a person with a launched machine and no human exposure is a capability that lowers the threshold for choosing to do so. The worry is not a single wrongful death. It is a quiet, steady expansion of who gets killed, because the act has become cheap enough that the bar for ordering it keeps dropping. Johnson’s wider argument about machine intelligence and escalation rests on exactly this mechanism: systems that compress time and remove human friction make it easier to act and harder to stop, and that is as true of a covert program as it is of a nuclear crisis.

The error problem deepens the concern rather than standing apart from it. The Lavender reporting documented an error rate of around ten percent, accepted in practice because the human review had shrunk to twenty seconds. A ten percent error rate in a high-volume war is a horrifying number. A ten percent error rate in a covert elimination program means that one in ten of the people quietly killed was not the person the state believed it was killing, and because covert operations are by design unaccountable, that error is never examined, never corrected, and never even acknowledged. The accountability vacuum that already surrounds clandestine killing becomes far more dangerous when the killing is performed by systems whose reasoning is opaque even to their operators. The legal and ethical debate over targeted killing has struggled for two decades to impose restraint on human covert programs. It is nowhere near ready for programs that no human fully directs.

There is a further moral hazard, subtler than error and harder to legislate against, and it concerns the diffusion of responsibility. When a covert killing is carried out by a human team, the chain of accountability, however hidden from the public, is at least intact inside the service. An officer chose the target, an officer approved the method, an officer ordered the act, and those officers know what they did. When the same killing is produced by a system, that chain frays at every link. The data scientist who built the targeting model did not select the victim. The analyst who approved the model’s recommendation in twenty seconds did not investigate it. The commander who authorized the program did not see the specific face. Each participant did something small, technical, and individually defensible, and the lethal outcome emerges from the combination without any single person having made the kind of weighty, deliberate choice that the killing of a human being morally demands. Responsibility does not vanish in such a system. It disperses, and a thing that is everyone’s responsibility in a thin slice is, in practice, no one’s responsibility in full. Covert programs already exploit the gap between what a state does and what it can be made to answer for. Automation widens that gap from the inside, because it lets the people running the program feel, with some sincerity, that the machine did it.

The asymmetry of the ethical debate is itself worth noticing. The case for automation is concrete, immediate, and easy to make in a briefing room: fewer officers at risk, faster results, lower cost, cleaner deniability, and a plausible claim of greater precision. The case against it is diffuse, deferred, and abstract: a slow erosion of thresholds, a hollowing of human judgment, a class of errors that will never be audited, a precedent that other states will copy. A decision-maker weighing those two columns is weighing a vivid set of gains against a set of harms that are real but will not appear on any quarterly review. Institutions are not built to give proper weight to that second column, and the honest forecast must record that the structural pressures all run one direction. The technologies will be adopted because the case for adopting them is the case that wins inside the kind of room where the choice gets made.

The constraint that ought to survive all of this is the principle of meaningful human control, the idea that a person must make and own the decision to take a specific life. It is the principle Scharre’s own work on Pentagon policy was built around, and it is the principle most at risk from the very efficiency these technologies promise. A system designed to compress the sensor-to-shooter timeline is, by its nature, a system designed to give the human less time and less reason to intervene. Holding the line on human control is not a technical problem. It is a choice that institutions have to keep making against the constant pull of their own tools toward speed, and the honest forecast is that the pull is strong and the line is thin.

The Race to Build Tomorrow’s Shadow War

These technologies are not arriving evenly. They are being built by a handful of states with deep defense-technology bases, and the pattern of who leads in each domain will shape which services hold the covert advantage through 2030.

The United States leads in the integration of artificial intelligence into targeting, with Project Maven as the most mature example of a system that turns sensor data into target recommendations at industrial speed, and it retains formidable depth across drones, cyber capability, and biometric research, much of it driven by programs whose outputs reach the broader research community. China is the clearest leader in the breadth of biometric surveillance, having built the densest network of facial-recognition infrastructure in the world and exported elements of it widely, and it is a serious force in drone manufacturing, with Chinese components and processing boards turning up inside the autonomous loitering munitions used in other countries’ wars. Israel sits at the front of practical, battle-tested application, with Unit 8200’s targeting systems, a long history of drone development from the early Harpy onward, and the Stuxnet precedent in cyber sabotage. Turkey has become an unexpectedly large player through Baykar and STM, whose drones and loitering munitions have spread across continents and now reach the land forces of NATO and European Union members. Russia and Ukraine, locked in the most intense drone war in history, have become the world’s fastest innovators in autonomous and AI-enabled uncrewed systems, with battlefield-driven development outpacing the slower laboratory work of larger powers.

India’s position in this race is the question that matters most for the future of its own campaign. New Delhi has demonstrated, through the drone exchanges of the 2025 conflict and the combat use of the S-400 air defense system, that it treats uncrewed and networked warfare as central rather than peripheral. Its technology sector is large, its software talent is among the world’s deepest, and its external intelligence service has shown across the shadow war that it can plan and execute with discipline. The raw ingredients of an algorithmic covert capability are present. The open question is integration: whether an intelligence culture that has succeeded through human tradecraft, careful planning, and the disciplined motorcycle method can absorb data science, autonomous systems, and biometric analysis into a single coherent doctrine, or whether it adopts the tools piecemeal and loses the discipline that made the campaign effective in the first place. The states that win this race will not be the ones with the best individual technologies. They will be the ones that fuse the four into a single working system, and that is an institutional achievement as much as a technical one.

There is a feature of this race that distinguishes it sharply from earlier technological contests, and it should temper any expectation that leadership will stay with the established powers. The components of an algorithmic covert capability are, to an unusual degree, commercial and diffusible. A small quadcopter is a consumer product. The processing boards that give a loitering munition its autonomy are sold for robotics and industrial vision. Facial-recognition software is a download. Targeting models can be trained by anyone with data, talent, and graphics processors, all three of which are widely available. The Libya episode that the United Nations panel documented is the warning: the autonomous machine in that case was a manufactured export, used far from the country that built it, by a party that did not invent any of the underlying science. The implication for the shadow war is uncomfortable. A capability that begins in the hands of advanced services does not stay there. Within a few years of maturing, a usable version of it is available to second-tier states, and a cruder version is available to the non-state groups that have always been participants in this conflict rather than spectators. The race is not only between major intelligence services. It is also a race between states and the proliferation curve itself, and the proliferation curve has no loyalty.

For India specifically, the race poses a question of institutional identity rather than mere capability. A service can buy drones, license recognition software, and hire data scientists without ever deciding what kind of service it intends to be. The harder choice is doctrinal. An intelligence culture that has built its reputation on patience, human tradecraft, and the disciplined restraint of the motorcycle method must decide whether the new tools will serve that culture or quietly replace it. The optimistic path is fusion under doctrine, where algorithmic surveillance, autonomous options, and biometric tracking are absorbed into an existing tradition of careful, accountable, human-directed action, making that tradition faster and sharper without changing its character. The pessimistic path is adoption without doctrine, where the tools are bought piecemeal because rivals have them, each one optimized for speed, and the cumulative effect is a service that has traded its discipline for its efficiency without ever having voted to do so. Which path New Delhi takes will not be announced. It will be revealed slowly, in the texture of operations across the coming decade.

Pakistan’s position in the same race determines how long the technological advantage lasts for whoever gains it first. A campaign of autonomous, algorithmic elimination is devastating against a defender who cannot answer it and merely difficult against a defender who can. If Pakistan’s military and its intelligence service invest in counter-drone systems, in feeding false data to biometric trackers, in hardening the digital surfaces around protected figures, and in the electronic warfare that has proven decisive in Ukraine, then the technological shadow war becomes a contest rather than a rout. The likeliest future is not one side achieving permanent dominance. It is an accelerating cycle of measure and counter-measure, the same dynamic that has governed where the world’s intelligence agencies stand against one another for as long as the trade has existed, now running at the speed of software.

Where the Forecast Breaks Down

A projection this confident has an obligation to mark its own weak points, because technology forecasting has a long and humbling record of error, and the four trends described here will not unfold as cleanly as a tidy argument makes them sound.

The first and largest caveat is the gap between laboratory capability and operational reliability, the gap that Scharre has spent a career insisting people respect. A facial-recognition system that performs beautifully on a benchmark dataset can fail badly against a real target in poor light, at an awkward angle, partially covered, in a crowd. An autonomous drone that navigates flawlessly in a test can be defeated by weather, by jamming it was not designed to expect, or by a coordination failure in a swarm. A targeting model validated on historical data can break against an adversary who has learned to feed it the inputs that make it wrong. The history of military technology is full of capabilities that worked in demonstration and disappointed in war. Every projection in this article should be read with that history in mind: the direction of travel is clear, the speed is not, and the assumption that a technology demonstrated in 2026 will be reliable in covert use by 2030 is an assumption, not a fact.

A related caveat concerns cost and scale, two factors that are usually assumed to favor the machine and sometimes do not. The intuitive case for automation rests on the machine being cheaper than the team, and at the level of a single elimination that may hold. At the level of a program it is far less certain. A credible algorithmic covert capability is not one drone and one model. It is a sustained pipeline of data collection, model training and retraining, secure infrastructure, specialized personnel, counter-counter-measures, and the constant maintenance that any complex system demands. The visible per-killing cost falls while the hidden institutional cost rises, and a service that adopts the tools expecting savings may find it has instead bought an expensive permanent dependency. There is also a scale trap. A method that is cheap enough to use often will be used often, and a program built for high volume generates a high volume of errors, of evidence, and of opportunities for exposure. The human method is slow and expensive in a way that quietly limits how much of it a service does. Remove that limit and the program may grow until its own scale becomes the vulnerability.

The forecast also underweights, by the nature of any forward projection, the speed of co-evolution. This article describes the hunting technologies maturing along a fairly smooth curve, but technological contests do not move smoothly. They move in jumps, as one side fields a capability and the other answers it with a counter that resets the problem. A breakthrough in counter-autonomy, a cheap and reliable way to spoof biometric systems at scale, a hardening technique that closes the cyber surface around protected figures, any of these would not merely slow the forecast. It would bend it. The shadow war has always been an arms race, and the distinguishing feature of an arms race is that no projection of one side’s capability survives contact with the other side’s response. The four technologies will mature. What they will be worth in 2030 depends on a defensive curve this article cannot see.

The second caveat is that adversaries adapt, and the forecast above can read, if one is not careful, as though only the hunting side gets to use the new tools. That is false. Every technology described here has a counter, and the counters are developing as fast as the capabilities. Counter-drone systems, jammers, and physical defenses degrade the autonomous machine. Biometric spoofing and the deliberate poisoning of databases degrade the tracker. Targeted men can be coached to live in ways that minimize their digital and biometric surface. A protective service that takes the threat seriously can blunt a great deal of it. The future is a contest between adapting opponents, not a one-sided demonstration of capability against a static target, and any forecast that forgets the defender is only half a forecast.

The third caveat is the deepest, and it concerns the human factors that no amount of capability erases. Covert operations have always failed less often on technology than on people, on politics, and on the unpredictable consequences of acting. An autonomous program can be exposed by a leak, by a captured machine analyzed back to its origin, by a defector, or by a single error that becomes an international incident. The diplomatic blowback that has trailed the human campaign does not disappear because the killing is automated; if anything, the discovery of an autonomous covert program might provoke a sharper international reaction than a motorcycle shooting ever did. The shift from human to machine changes the texture of covert work. It does not repeal the politics, the law, the risk of escalation, or the simple fact that hunting people across another state’s territory remains an act with consequences no technology can engineer away. The forecast describes how the killing will be done. It cannot promise that doing it will be wise.

What the Forecast Teaches

Set the four technologies side by side and the lesson is not really about artificial intelligence, drones, code, or biometrics as separate marvels. The lesson is about a single direction of travel, and a single question that direction forces every intelligence service to answer.

The direction is the steady removal of the human being from the act of clandestine killing. The surveillance team becomes an algorithm. The shooter becomes a machine or a piece of software. The witness sees a drone or sees nothing. Across a decade, the trade is moving from a craft performed by trained people toward a process executed by systems, and the motorcycle method that has defined India’s campaign is best understood, from the vantage point of 2030, as the final mature form of the human era of covert killing. It is disciplined, it is effective, and it is human in every part, and that combination is exactly what the next decade is engineering away.

Whether the motorcycle becomes a museum piece depends on which target one has in mind. Against a hardened, low-technology figure living a deliberately analog life in a Pakistani town, the human method will persist longest, because such a man presents almost no cyber surface, may evade aerial recognition, and can sometimes stay ahead of the databases. Against the modern, connected, mobile target, the technological methods will arrive first and dominate soonest. The likeliest future through 2030 is therefore not the replacement of the motorcycle but the layering of capabilities on top of it: algorithmic surveillance feeding human teams, drones used where drones fit, cyber options held for the targets that present a surface, biometric tracking running underneath all of it as the system that makes every other method faster. The motorcycle does not vanish. It becomes one tool among several, and a steadily less central one.

It is worth being precise about what is gained and what is lost in that transition, because the loss is easy to romanticize and the gain is easy to oversell. Nothing about the human method is morally superior simply because it is human. A motorcycle killing is still a killing, and the discipline that has characterized India’s campaign was never a property of the motorcycle. It was a property of the officers who planned each operation, weighed each target, and accepted the friction of doing the work carefully. Restraint, in other words, has been a human product, manufactured by people making deliberate choices under real constraints. The danger of the coming decade is not that machines are crueler than people. It is that machines are indifferent, and indifference at speed and scale removes the very friction inside which human restraint was produced. The lesson the forecast teaches is therefore not nostalgia for an older method. It is a warning that the quality the campaign has been praised for is not embedded in any technology and will not survive on its own. If it is to continue, it has to be deliberately rebuilt into the new system, designed in as a requirement rather than assumed as a habit, because the tools themselves will not carry it forward.

The question the forecast leaves on the table is the one that should outlast every specific prediction in it. The technologies will do what their physics allows; that part is close to settled. What is not settled is whether the states that hold them will keep a human being meaningfully in command of the decision to end a specific life, or whether the relentless pull of speed, scale, and deniability will hollow that decision out until it is a twenty-second review of a machine’s recommendation. India’s shadow war has so far been notable for its discipline, and discipline has been a human achievement, a matter of patient officers making careful choices. The deepest risk of the coming decade is not that the machines will fail. It is that they will succeed so well that the discipline becomes optional, and a campaign that was once a series of deliberate human decisions becomes an automated function that simply runs. What the next phase of India’s counter-terror doctrine chooses to do with that risk, more than any single technology, will determine what the shadow war becomes.

Frequently Asked Questions

Q: How will artificial intelligence change covert operations?

Artificial intelligence attacks the most labor-intensive stage of a covert elimination, which is the surveillance that establishes a target’s identity, location, and routine. Systems already in military use, such as Israel’s Lavender and the American Project Maven, convert vast volumes of intercepted data into ranked lists of human targets at a speed no team of analysts can match. For a selective campaign the value is certainty rather than volume: a model tuned to monitor one known individual across every digital surface he touches could replace weeks of exposed human watching with continuous algorithmic tracking. The cost is that the human officer shrinks from investigator to reviewer, with less time and less context to catch the moment the machine is wrong.

Q: Could autonomous drones replace human assassins?

For many targets, yes, and the technology to do it largely exists. Loitering munitions such as Turkey’s Kargu are battle-tested and exported widely, and a United Nations panel reported a Kargu-2 attacking targets in Libya with no data link to an operator. Russia’s V2U loitering munition has been documented choosing targets with onboard artificial intelligence and navigating without satellite signals. A bird-sized machine that carries a target’s face in its memory, launches from kilometres away, and strikes on its own removes the shooter, the witness, and the escape problem all at once. The limits are payload, weather, and the difficulty of facial recognition from a moving aerial platform, which means the human method will persist alongside drones rather than vanishing immediately.

Q: What is cyber assassination and is it possible?

Cyber assassination means causing a death by manipulating the networked systems a person depends on rather than by using a weapon against them directly. The proof that code can cause physical destruction is Stuxnet, the software discovered in 2010 that made Iranian nuclear centrifuges destroy themselves while reporting normal readings. Extending that logic to a person is conceivable through a sabotaged vehicle control system, a reprogrammed medical device, or a manipulated building system, each of which would read to investigators as an accident. The method is real but limited, because it requires the target to use vulnerable connected technology, and many hardened militants live deliberately low-technology lives that present almost no cyber surface.

Q: How will biometric tracking affect intelligence operations?

Biometric tracking attacks the oldest survival strategy of a hunted person, which is changing identity to disappear. A face, an iris, a voice, and a gait cannot be altered by adopting a new name, and once a target’s biometric signature is enrolled in a searchable database, the hunt inverts: the campaign no longer has to find where a man hides, it only has to wait for him to pass a sensor. Border systems already capture faces, fingerprints, and iris scans routinely, and research systems now fuse face, body shape, and gait to identify people at long range and from elevated viewpoints. This is also the technology with the gravest civil-liberties implications, because the same systems expose journalists and dissidents as readily as wanted militants.

Q: Will India’s motorcycle method become obsolete?

Not entirely, and not soon, but its centrality will fade. The motorcycle-borne shooting is the mature final form of the human era of covert killing, and it will persist longest against hardened, low-technology targets who present little cyber surface and can sometimes evade aerial recognition. Against modern, connected, mobile targets the technological methods will arrive first. The realistic future through 2030 is layering rather than replacement: algorithmic surveillance feeding human teams, drones used where they fit, cyber options held for connected targets, and biometric tracking running underneath all of it. The motorcycle becomes one tool among several rather than the signature of the entire campaign.

Q: Which countries lead in covert-operations technology?

Leadership is split by domain. The United States leads in integrating artificial intelligence into targeting through programs like Project Maven and retains depth across drones, cyber capability, and biometric research. China leads in the breadth of biometric surveillance infrastructure and is a major drone manufacturer. Israel leads in practical, battle-tested application across AI targeting, drones, and cyber sabotage. Turkey has become a large exporter of drones and loitering munitions. Russia and Ukraine are the fastest innovators in autonomous uncrewed systems, driven by the most intense drone war in history. The decisive advantage will go to whichever service best fuses the separate technologies into one coherent doctrine.

Q: Does technology make state killing too easy?

This is the central ethical worry, and it is the one James Johnson has pressed hardest. Every element of friction in the current method, the cost of training a team, the risk to that team, the political weight of ordering people into danger, acts as a brake that forces a decision-maker to want a killing badly enough to pay a real price for it. Removing the human team removes the friction and collapses the price. A capability that lets a state eliminate a person with a launched machine and no human exposure lowers the threshold for choosing to do so, and the danger is not a single wrongful death but a quiet, steady expansion of who gets killed because the act has become cheap.

Q: What ethical constraints should apply to AI-driven targeting?

The constraint that matters most is meaningful human control, the principle that a person must make and own the decision to take a specific life. It is the principle that shaped early Pentagon policy on autonomy in weapons, and it is the principle most at risk from the very efficiency these systems promise, because a tool designed to compress the time between detecting a target and striking it is by nature a tool that gives the human reviewer less room to intervene. Holding that line is not a technical problem but an institutional choice that has to be remade constantly against the pull of speed, scale, and deniability.

Q: What was Stuxnet and why does it matter for the future?

Stuxnet was malicious software discovered in 2010, widely attributed to a joint American and Israeli program, that attacked the centrifuges enriching uranium at Iran’s Natanz facility. It altered the speed of the centrifuges to make them tear themselves apart while reporting normal readings to the engineers watching the controls. Stuxnet matters for the future of covert work because it established two principles: software can produce physical destruction, and it can disguise that destruction as ordinary mechanical failure. Those are exactly the properties that would make a cyber operation against a person attractive to an intelligence service, because the result would read as an accident rather than a killing.

Q: What is a loitering munition?

A loitering munition is a small uncrewed aircraft that carries an explosive charge and is designed to orbit an area until a target is identified, then dive into it and detonate. It sits between a missile and a drone, combining the persistence of a surveillance aircraft with the lethality of a guided weapon. The war in Ukraine turned loitering munitions into a defining battlefield tool, and the most significant recent development is autonomy: machines like the Russian V2U can now search for and select targets with onboard artificial intelligence and navigate without satellite signals, which makes them difficult to jam and impossible to trace back through a radio link to an operator.

Q: How does facial recognition threaten covert operatives?

Facial recognition threatens both the field officer and the target. For the field officer, the dense facial-recognition networks built into borders, airports, and city surveillance make it far harder to move through hostile territory under a false identity without being matched against a database. For the target, facial recognition is the engine that strips away the anonymity a wanted man relies on, because a face has a fixed geometry that a new name cannot change. Combined with iris and gait data, a facial-recognition match turns the discovery of a hidden person from a months-long human investigation into a query that resolves the moment the person passes a camera.

Q: What is gait recognition and why is it hard to defeat?

Gait recognition identifies a person by the unique rhythm and mechanics of the way they walk. It is hard to defeat because, unlike a face, a gait cannot easily be covered, disguised, or consciously changed for long. Researchers value it precisely because it works at the long distances and from the elevated viewpoints that covert surveillance and aerial platforms operate at, and where ordinary facial recognition struggles. The most advanced systems now fuse gait with face and body shape into a single identification designed to function in poor light, at a distance, and against a partially obscured subject, which makes it a particularly serious threat to anyone whose survival depends on not being recognized.

Q: Did an autonomous drone ever kill someone without human authorization?

It may have. A United Nations panel of experts examining the Libyan civil war reported that a Turkish-made Kargu-2 loitering munition was used to attack targets without a data link between the operator and the machine, describing it as a true fire, forget, and find capability. If the panel’s account is accurate, that incident may represent the first time a machine selected and attacked a human being with no human confirming the specific strike. The manufacturer maintains that the system operates under a man-in-the-loop principle, and the episode remains contested, but it illustrates how thin the line between supervised and fully autonomous lethal action has already become.

Q: How accurate are AI targeting systems today?

Accurate enough to be used, and not accurate enough to be trusted blindly. The reporting on Israel’s Lavender system documented an error margin of roughly ten percent, meaning the system misidentified targets a meaningful fraction of the time. In a high-volume conventional war that number is treated as tolerable. In a covert elimination program it is alarming, because a ten percent error rate means one in ten quietly killed people was not the intended target, and the unaccountable nature of covert work means that error is never examined or corrected. The accuracy problem is not a reason the technology will not be adopted; it is a reason its adoption is dangerous.

Q: Could India build an AI targeting system of its own?

The ingredients are present. India has one of the world’s deepest pools of software talent, a large technology sector, and an external intelligence service that has shown across the shadow war that it can plan and execute with discipline. The combat use of drones and the S-400 in the 2025 conflict demonstrated that New Delhi treats networked and uncrewed warfare as central. The open question is integration rather than capability: whether an intelligence culture built on human tradecraft can absorb data science, autonomous systems, and biometric analysis into one coherent doctrine without losing the discipline that made the campaign effective. Building the components is achievable. Fusing them into a working system is the harder institutional task.

Q: What is the sensor-to-shooter timeline?

The sensor-to-shooter timeline is the military term for the time that elapses between a sensor first detecting a potential target and a weapon striking it. Shortening that timeline is the central organizing goal of modern military artificial intelligence, because a faster timeline means more targets struck and less chance for a target to move. Project Maven illustrates the trend: an operator using the system can reportedly sign off on far more targets per hour than without it. The concept matters for covert work because compressing the timeline also compresses the human decision, leaving the reviewer less time and less context to question whether a strike should happen at all.

Q: Will technology make covert operations easier to attribute or harder?

It depends on the technology. A cyber operation, done well, makes attribution far harder, because a death engineered through a sabotaged system can read as an accident with no projectile, no wreckage, and no obvious violence at all. An autonomous drone is more ambiguous: it removes the human witness and the capturable shooter, but a recovered machine can sometimes be analyzed back toward its origin. On balance the technologies push toward deniability, since their shared appeal is the removal of the human operative whose capture or testimony is the most common way a covert program is exposed. The discovery of an autonomous program, however, could provoke a sharper international reaction than a conventional shooting.

Q: How might Pakistan defend its protected figures against these technologies?

Every technology described here has a counter, and a protective service that takes the threat seriously can blunt a great deal of it. Counter-drone systems, signal jammers, and physical defenses such as netting degrade autonomous machines, as the war in Ukraine has shown. Biometric trackers can be defeated by feeding databases false data and by coaching protected figures to minimize their biometric and digital surface. Hardening the connected systems around a high-value individual reduces the cyber surface. Electronic warfare, which has proven decisive in Ukraine, is effective against uncrewed systems generally. The likely result is not one-sided dominance but an accelerating cycle of measure and counter-measure.

Q: Is there any international law governing autonomous weapons?

International humanitarian law, as codified in the Geneva framework, requires that combatants distinguish between fighters and civilians and that any civilian harm be proportionate to military necessity. Critics of autonomous weapons argue that only humans can reliably make those fine distinctions in the chaos of a real situation, and there have been sustained international efforts to negotiate limits or a ban on weapons that select and engage targets without human control. No comprehensive binding treaty specific to lethal autonomous weapons has been concluded, and covert programs operate in deniability that makes any legal framework even harder to enforce. The governance of these systems lags well behind their capability.

Q: What happens to deniability when machines do the killing?

Deniability arguably increases, because the oldest way a covert program is exposed is through its people: an operative who is followed, a shooter who is captured and interrogated, a witness who describes a face. Removing the human element removes those failure points. A cyber operation can leave a death that is never even classified as a killing. An autonomous drone leaves no shooter and no witness who saw a human. The trade-off is that machines and code carry their own evidentiary traces, a recovered drone or a forensic analysis of malicious software, and the discovery of a deliberately automated killing program could itself become a scandal larger than any single operation. Deniability shifts from protecting the field officer to protecting the system.

Q: Could non-state groups acquire these covert technologies?

Yes, and this is one of the most destabilizing features of the shift. Unlike a stealth aircraft or a nuclear program, the components of an algorithmic covert capability are largely commercial. Small drones are consumer goods, vision-processing boards are sold for robotics, facial-recognition software is widely available, and targeting models can be trained by anyone with data and graphics processors. The United Nations panel that examined the 2020 Libya episode described an autonomous loitering munition used by a non-state party far from where it was manufactured. Militant organizations have always been participants in this conflict rather than bystanders, and the proliferation curve means a cruder version of every capability described here reaches them within a few years of it maturing. The technological shadow war will not stay a contest between states.

Q: Can an autonomous drone be traced back to the country that launched it?

Sometimes, and that uncertainty is part of its appeal to the launching service. A recovered or downed drone can be forensically analyzed, and its components, processing boards, software, and manufacturing marks can suggest an origin, though widely traded commercial parts make a confident attribution difficult. The harder problem is the gap between identifying where a machine was built and proving who ordered it into another state’s territory. Components from one country may be assembled in a second and deployed by a third. For a service that wants plausible deniability, a machine built from globally sourced commercial parts offers exactly the ambiguity that a captured human operative never could. Attribution becomes a matter of probability and inference rather than confession.

Q: What does meaningful human control actually require in practice?

It is more demanding than the phrase suggests. Meaningful human control is not satisfied by a person being technically present in the decision loop. A human who reviews a machine’s recommendation for twenty seconds, with no practical ability to investigate it and no real expectation of overriding it, is present but not in control. Genuine meaningful human control requires that the person have enough time, enough information, and enough institutional freedom to reach an independent judgment and to say no without penalty. It requires that the human decision be the operative one and the machine’s output be advice. The reason the principle is fragile is that every efficiency gain these technologies offer comes specifically from compressing the time and reducing the friction that meaningful control depends on. The principle and the speed are in direct tension.

Q: How has the war in Ukraine changed covert-operations technology?

The conflict has functioned as the largest live laboratory for uncrewed and autonomous systems in history, and its effects reach well beyond conventional warfare. It has driven drone development at a speed that peacetime laboratories cannot match, normalized first-person-view strike drones and loitering munitions, and pushed both sides toward autonomy as a direct response to electronic jamming. Terrain-matching navigation, onboard target recognition, and AI-assisted final approach were accelerated because jamming made human-piloted machines unreliable. Those same capabilities, matured under battlefield pressure, are exactly the ones that make a micro-drone viable as a covert tool. The war also demonstrated counter-drone measures at scale, which means the defensive half of the contest advanced just as fast. Covert planners worldwide have been studying both halves closely.

Q: Will these technologies make the shadow war more or less violent?

The honest answer is that it depends on which pressure dominates, and the two pressures point in opposite directions. The precision argument suggests less violence: better tracking and more discriminating delivery could mean fewer bystander deaths per operation than a firefight or a bomb in a crowded place. The threshold argument suggests more violence: by removing the cost, the risk, and the human friction that currently restrain how often a state chooses to kill, the technologies could expand the number of people targeted even as each individual operation becomes cleaner. It is entirely possible for both to be true at once, producing a future of more frequent killings that are each individually more precise. Whether the net effect is more or less violence is not a technical question. It is a question of restraint, and restraint is a choice rather than a property of the tools.