What’s the biggest thing you think you can take in a fight? According to a YouGov poll, 6% of Americans reckoned they could beat a grizzly bear bare-handed. But lest you take this as evidence of a, let’s say, optimistic national spirit, only 72% thought they could take a rat. I think I could win that fight.
What’s the smallest thing that could take you? While I like to think that, if cornered, I could take all manner of small rodent, I’m not sure even the brave 6% would take a 10-gram bullet to the brain. They also probably couldn’t win against 300 mg of cyanide, 10 mg of sarin gas, or just 0.1 μg of botulinum toxin, the deadliest known toxin. Numbers this small can be hard to grasp. All three substances are deadly, but the lethal dose of cyanide is more than a million times greater than that of botulinum toxin. A grizzly bear, by contrast, is merely a thousand times heavier than a rat.
We are still not close to the most dangerous object of all, pound for pound. A single smallpox virion weighs less than 10⁻¹⁴ grams, less than a millionth the mass of a lethal dose of botulinum. And yet, if you were unlucky enough to inhale even a few dozen virions and had not been vaccinated, you would probably be infected; absent treatment, you face a case fatality rate of around 30%.
I don’t need to explain to you how bears or bullets kill people. Cyanide is more selectively destructive. Each molecule disables one copy of a key protein, cytochrome c oxidase, and a lethal dose is simply enough cyanide molecules to disable enough of them. Botulinum toxin is more lethal still because it works catalytically: each molecule can destroy countless SNARE proteins, which neurons need to fire signals at muscles. A single botulinum toxin molecule can knock out an entire nerve terminal. With enough molecules to knock out enough terminals, you can no longer breathe.
Unlike these other examples, a single smallpox virion does not immediately harm you; given enough time, however, it could kill you. The trick is self-replication: when a virion infects a cell, it turns it into a machine for making more copies of itself. That cell bursts and dies, releasing hundreds of new virions into your body; those in turn infect new cells and make new virions of their own, and soon that single virion has become billions.
It gets worse. You can’t catch botulism or cyanide poisoning from someone else, and even the bear will eventually get tired. Smallpox shows no such restraint. As it multiplies, you begin to shed it, coughing and breathing out thousands of infectious particles; anyone nearby who inhales them is infected in turn, and one case becomes two, two become four, and soon you have an exponentially growing epidemic. We are all, of course, familiar with this from COVID-19: a comparatively mild virus that still killed millions of people and did tens of trillions of dollars of damage to the world economy. Smallpox killed hundreds of millions over the course of its existence before we finally eradicated it with vaccines.
Smaller, cheaper, scarier
Humans have attempted to weaponize essentially everything under the sun, from sticks and stones to K-pop. But biological systems are almost unique in their ability to physically self-replicate[1], and this makes biological weapons uniquely terrifying.
A few crude calculations are enough to make the point. For each weapon in the table below, conventional and unconventional, I estimate how many people it could kill under ideal conditions against a dense, undefended population, what it would cost a well-supplied state to produce, and how much it weighs. Dividing by the death toll gives a cost per death and a mass per death, a rough measure of how destructive each weapon really is. Methodological details are in the appendix; italics mark more speculative systems that have not yet been built.
In theory, a single virion could initiate a pandemic that kills a substantial fraction of humanity. Just how hard it is to build such a virion remains very much an open question. The cost used in this table is that of a focused R&D effort by a well-resourced team over several years, and whether that is a reasonable estimate given current or near-future technology is unclear to me. Unfortunately, there is really no in-principle barrier to creating a virus once a suitable design is known. In 2018, a research team assembled horsepox—a large poxvirus and close relative of smallpox—from mail-order DNA fragments for around $100,000. As viral reconstructions go, horsepox is a hard case: poxviruses have large, unwieldy genomes, and most other viruses are considerably smaller and easier to build. A commercial lab such as VectorBuilder will turn a known sequence into live virus particles for a few thousand dollars.
Pandemic pathogens are not the only threat we could face. Mirror bacteria, not shown in the table, are hypothetical organisms whose molecules have reversed chirality—mirror images of the ones life actually uses. That reversal would greatly reduce their visibility to the predators and immune systems that evolved to handle ordinary chirality. They could replicate both in the environment and within human bodies, and so could be very dangerous. Fortunately, creating mirror bacteria remains out of reach of current science. That may not always be the case, and scientific progress may facilitate the creation of other persistent environmental threats.[2]
Self-replication is so powerful that even traditional, non-contagious bioweapons like anthrax can be more lethal, per cost, than nuclear weapons. For military planners, the lack of transmission is not necessarily a bad thing; uncontrollable, indiscriminate destruction is not usually desirable even in a weapon. But compared to nuclear weapons, non-contagious biological weapons are generally seen as less reliable and more easily countered; this is probably why states have historically preferred the former.
The rest of the arsenal
Nuclear weapons are really the only thing that comes close to biological weapons. They couldn’t enter the “smallest things that can kill you” competition: nuclear weapons require a critical mass of uranium or plutonium to sustain a chain reaction, and this cannot be shrunk past a certain point. Bigger nuclear weapons can of course be built, but they are commensurately more expensive, and the trend line has in fact been toward smaller ones. Fortunately, physics doesn’t seem to allow for anything more exotic either (excepting, perhaps, false vacuum decay).
Chemical weapons, by contrast, are several orders of magnitude less destructive than nuclear weapons. Their designation as weapons of mass destruction is really only a historical accident; they aren’t even clearly more lethal than conventional munitions. One suspects the real reason they were banned was that they were never very useful for advanced militaries.
As AI improves, drones only get cheaper, more autonomous, and easier to field in large numbers. In the right circumstances they are already a cheaper way to deliver conventional explosives than a missile or an aircraft. But even with major engineering gains they would stay far less lethal than nuclear or biological weapons: a rifle round costs $0.20, so to beat a nuke’s $2 a death, a drone would have to be nearly free and kill with fewer than ten rounds.
Advanced AI could push economic growth and physical production far beyond historic precedents—and with them, drone armies at a scale we’ve never seen. That is worth worrying about, but it is not the sort of thing that could happen in secret. If a bad actor ends up in control of a large drone army, it’s because we either let them build it or, worse, built it for them.
What can we do?
Biological weapons, then, come out of this comparison looking worse than other weapons. But it is not all bad news. Nuclear weapons are really hard to defend against; the current state of the art consists of threatening the other guy with a nuclear weapon of your own. With biological weapons, fortunately, we have better options.
For a start, we can keep them hard to build. Good enough technology might one day make a pandemic pathogen cheap and easy to produce, but we are not there yet; the first thing we can do is keep it that way, raising the bar to building one rather than letting it fall, for example by screening the synthetic DNA an engineered pathogen has to be assembled from. And if one is built and released anyway, the next thing is to catch it early, for instance by sequencing what shows up in clinics and in wastewater to flag a new pathogen before it spreads.
Second, for all the menace of self-replication, a pathogen still has to get inside you before it can do anything, and there are only so many ways in. It is also a living thing, and living things can be killed. Air filters, ultraviolet light, chemical disinfectants, and respirators could help defend us from even the worst possible threats.
Finally, because a pathogen has to replicate inside us, we can try to use drugs or other medical countermeasures to kill it before it kills us. This is, unfortunately, easier said than done; we still don’t have effective treatments for many natural diseases, and millions still die each year in poor countries due to diseases that are easily curable. But even if defensive technologies lag, there is still a lot we could do to be better prepared.
Appendix: Lethality estimates
All entries are order-of-magnitude estimates. The maximum deaths column assumes ideal conditions against a dense, undefended population. Costs are marginal procurement and operations for a well-supplied state, with reusable platforms amortized over their service life. R&D costs are likewise assumed amortized, with the exception of the pandemic pathogen, which can only be used once. The table covers only weapon systems that don’t require a human operator in the field. Hand-wielded weapons are cheap—a machete costs O($10) and can be used many times—but their lethality is limited by the operator, who is expensive and placed at high risk.
Pandemic pathogen: The maximum death count of one billion is a round figure for a highly transmissible, highly lethal agent that spreads globally. The cost, and its large uncertainties, are discussed in the main text. The estimate of 10⁻¹² g for a single infectious dose is conservative. Individual virions weigh 10⁻¹⁷–10⁻¹⁵ g, and infectious doses range from O(10) to well over 10⁶ virions.
Nuclear weapons: A standard 1 Mt thermonuclear warhead costs roughly $2.5M and weighs roughly 1,100 kg—the mass of the B83, the closest US example at 1.2 Mt. Delivered by airplane, a 1 Mt airburst over a dense city kills roughly 1.5 million (NUKEMAP); the amortized cost of the aircraft is negligible, giving a cost per death of roughly $2. A Trident D5 carries 8 W88 warheads, each 475 kt; spread across cities, the full payload could kill perhaps 4 million. The loaded missile costs roughly $80M and weighs 60 t.
Anthrax, by airplane: An aircraft spray of 100 kg of anthrax spores over a large city could kill 1–3 million (1993 OTA report); the table uses the midpoint of this range. The cost of producing and delivering the agent by light aircraft is roughly $200k (inflation-adjusted from the OTA-era estimate).
Anthrax, by Scud: A Scud carrying 30 kg of anthrax spores could kill 30,000–100,000 (OTA report); the table uses the midpoint of 50,000. A Scud-B costs roughly $1M (Zaloga) and weighs 5,900 kg, which dominates the total cost.
Conventional munitions, by airplane: A B-52 carries about 30 t of ordnance—roughly 130 Mk 82 (500 lb) bombs at $2,500 each. A 500 lb bomb has a lethal radius against unprotected personnel of roughly 50 m, for a lethal area of about 8,000 m² per bomb; 130 bombs cover a total lethal area of roughly 1 km². In a dense urban area (10,000 people per km²), this puts roughly 10,000 people in the nominal lethal area. Buildings provide substantial cover; as a rough upper bound, 1,000 deaths is plausible. For comparison, the Tokyo firebombing of March 1945 killed 100,000 with 1,665 tons of incendiaries at a density of 25,000/km²—about 60 deaths per ton—but incendiaries are far more destructive than high explosive against flammable cities like wartime Tokyo, which was largely built of wood. A 6-hour sortie costs $420k in flight time ($70k per flight hour) plus $300k in munitions, for a total of about $700k.
Conventional munitions, by Scud: The best data on conventional ballistic missile lethality comes from Lewis, Fetter, and Gronlund (1993), who use V-2 casualty records from London (518 impacts, 2,510 deaths) to build an empirical model. The V-2 warhead is comparable in size to the Scud-B (980 kg each). In London, the average death radius was only 21 m, giving a lethal area of 1,300 m² per impact. Scaled to the Scud-B’s somewhat larger energy release and a population density of 10,000/km², the model predicts roughly 20 deaths per strike—consistent with the Iran-Iraq War of the Cities, where 533 Iraqi missiles killed 2,312 Iranian civilians (about 4 per strike with sheltering). Scud costs of $1M and mass of 5,900 kg are the same as for anthrax.
Sarin nerve agent, by airplane: The OTA report and Lord Lyell (1996) estimate that a light aircraft dispersing 1,000 kg of sarin over a city kills 400–800—no better than the same aircraft dropping an equivalent weight of high-explosive bombs. A chemical sortie costs $200k.
Sarin nerve agent, by Scud: The same sources estimate 60–100 deaths from a Scud carrying 300 kg of sarin—modestly more than a conventional Scud warhead (20 deaths), but not by an order of magnitude. Chemical lethality is highly weather-dependent; these figures assume ideal conditions for the attacker.
Hobby drone with grenade: A consumer FPV drone dropping a grenade could kill maybe two people. Ukrainian frontline FPV drones cost roughly $400–500 per unit. The total mass is about 1.5 kg.
TIKAD gun drone: The TIKAD is an armed octocopter mounting a rifle or grenade launcher, killing roughly 10 per mission against undefended targets. No public unit price exists; the $6,000 per-mission figure assumes a $120k platform amortized over 20 missions, which is an order-of-magnitude estimate based on comparable systems. The drone weighs 2.5 kg.
QinetiQ MAARS ground robot: The MAARS is a remotely operated ground robot armed with a 7.62mm machine gun or 40mm grenade launcher, killing roughly 100 per mission with a full ammunition load. No public price exists; the $17,500 per-mission figure assumes a $350k platform (inferred from the comparable TALON/SWORDS at $230k) amortized over 20 missions. The robot weighs 8.5 kg.
MQ-9 Reaper: Armed with four Hellfire missiles ($115–150k each), a single sortie kills roughly 50. The marginal sortie cost is roughly $700,000—the four missiles plus operating costs—with the $30M airframe treated as sunk and amortized over its service life. The payload and fuel weigh roughly 1,400 kg.
Kinetic orbital bombardment: A 5-tonne tungsten rod delivers only tens of tonnes of TNT-equivalent on impact (discussion), so the 5,000 death toll requires a direct hit on a dense district. The launch alone costs roughly $5M at current commercial rates ($1,000/kg to orbit); the delivery platform adds roughly $10M, for a total of about $15M.
Some of the most dangerous non-physical entities also self-replicate. Computer worms spread between machines, dangerous ideas—from multilevel marketing schemes to genocidal ideologies—between minds, and a capable enough AI could copy itself across hardware. ↩︎
Nanotechnological “grey-goo” is an extreme example, though it would face resource constraints and vulnerabilities similar to those of biological organisms. ↩︎
What’s the biggest thing you think you can take in a fight? According to a YouGov poll, 6% of Americans reckoned they could beat a grizzly bear bare-handed. But lest you take this as evidence of a, let’s say, optimistic national spirit, only 72% thought they could take a rat. I think I could win that fight.
What’s the smallest thing that could take you? While I like to think that, if cornered, I could take all manner of small rodent, I’m not sure even the brave 6% would take a 10-gram bullet to the brain. They also probably couldn’t win against 300 mg of cyanide, 10 mg of sarin gas, or just 0.1 μg of botulinum toxin, the deadliest known toxin. Numbers this small can be hard to grasp. All three substances are deadly, but the lethal dose of cyanide is more than a million times greater than that of botulinum toxin. A grizzly bear, by contrast, is merely a thousand times heavier than a rat.
We are still not close to the most dangerous object of all, pound for pound. A single smallpox virion weighs less than 10⁻¹⁴ grams, less than a millionth the mass of a lethal dose of botulinum. And yet, if you were unlucky enough to inhale even a few dozen virions and had not been vaccinated, you would probably be infected; absent treatment, you face a case fatality rate of around 30%.
I don’t need to explain to you how bears or bullets kill people. Cyanide is more selectively destructive. Each molecule disables one copy of a key protein, cytochrome c oxidase, and a lethal dose is simply enough cyanide molecules to disable enough of them. Botulinum toxin is more lethal still because it works catalytically: each molecule can destroy countless SNARE proteins, which neurons need to fire signals at muscles. A single botulinum toxin molecule can knock out an entire nerve terminal. With enough molecules to knock out enough terminals, you can no longer breathe.
Unlike these other examples, a single smallpox virion does not immediately harm you; given enough time, however, it could kill you. The trick is self-replication: when a virion infects a cell, it turns it into a machine for making more copies of itself. That cell bursts and dies, releasing hundreds of new virions into your body; those in turn infect new cells and make new virions of their own, and soon that single virion has become billions.
It gets worse. You can’t catch botulism or cyanide poisoning from someone else, and even the bear will eventually get tired. Smallpox shows no such restraint. As it multiplies, you begin to shed it, coughing and breathing out thousands of infectious particles; anyone nearby who inhales them is infected in turn, and one case becomes two, two become four, and soon you have an exponentially growing epidemic. We are all, of course, familiar with this from COVID-19: a comparatively mild virus that still killed millions of people and did tens of trillions of dollars of damage to the world economy. Smallpox killed hundreds of millions over the course of its existence before we finally eradicated it with vaccines.
Smaller, cheaper, scarier
Humans have attempted to weaponize essentially everything under the sun, from sticks and stones to K-pop. But biological systems are almost unique in their ability to physically self-replicate [1] , and this makes biological weapons uniquely terrifying.
A few crude calculations are enough to make the point. For each weapon in the table below, conventional and unconventional, I estimate how many people it could kill under ideal conditions against a dense, undefended population, what it would cost a well-supplied state to produce, and how much it weighs. Dividing by the death toll gives a cost per death and a mass per death, a rough measure of how destructive each weapon really is. Methodological details are in the appendix; italics mark more speculative systems that have not yet been built.
In theory, a single virion could initiate a pandemic that kills a substantial fraction of humanity. Just how hard it is to build such a virion remains very much an open question. The cost used in this table is that of a focused R&D effort by a well-resourced team over several years, and whether that is a reasonable estimate given current or near-future technology is unclear to me. Unfortunately, there is really no in-principle barrier to creating a virus once a suitable design is known. In 2018, a research team assembled horsepox—a large poxvirus and close relative of smallpox—from mail-order DNA fragments for around $100,000. As viral reconstructions go, horsepox is a hard case: poxviruses have large, unwieldy genomes, and most other viruses are considerably smaller and easier to build. A commercial lab such as VectorBuilder will turn a known sequence into live virus particles for a few thousand dollars.
Pandemic pathogens are not the only threat we could face. Mirror bacteria, not shown in the table, are hypothetical organisms whose molecules have reversed chirality—mirror images of the ones life actually uses. That reversal would greatly reduce their visibility to the predators and immune systems that evolved to handle ordinary chirality. They could replicate both in the environment and within human bodies, and so could be very dangerous. Fortunately, creating mirror bacteria remains out of reach of current science. That may not always be the case, and scientific progress may facilitate the creation of other persistent environmental threats. [2]
Self-replication is so powerful that even traditional, non-contagious bioweapons like anthrax can be more lethal, per cost, than nuclear weapons. For military planners, the lack of transmission is not necessarily a bad thing; uncontrollable, indiscriminate destruction is not usually desirable even in a weapon. But compared to nuclear weapons, non-contagious biological weapons are generally seen as less reliable and more easily countered; this is probably why states have historically preferred the former.
The rest of the arsenal
Nuclear weapons are really the only thing that comes close to biological weapons. They couldn’t enter the “smallest things that can kill you” competition: nuclear weapons require a critical mass of uranium or plutonium to sustain a chain reaction, and this cannot be shrunk past a certain point. Bigger nuclear weapons can of course be built, but they are commensurately more expensive, and the trend line has in fact been toward smaller ones. Fortunately, physics doesn’t seem to allow for anything more exotic either (excepting, perhaps, false vacuum decay).
Chemical weapons, by contrast, are several orders of magnitude less destructive than nuclear weapons. Their designation as weapons of mass destruction is really only a historical accident; they aren’t even clearly more lethal than conventional munitions. One suspects the real reason they were banned was that they were never very useful for advanced militaries.
As AI improves, drones only get cheaper, more autonomous, and easier to field in large numbers. In the right circumstances they are already a cheaper way to deliver conventional explosives than a missile or an aircraft. But even with major engineering gains they would stay far less lethal than nuclear or biological weapons: a rifle round costs $0.20, so to beat a nuke’s $2 a death, a drone would have to be nearly free and kill with fewer than ten rounds.
Advanced AI could push economic growth and physical production far beyond historic precedents—and with them, drone armies at a scale we’ve never seen. That is worth worrying about, but it is not the sort of thing that could happen in secret. If a bad actor ends up in control of a large drone army, it’s because we either let them build it or, worse, built it for them.
What can we do?
Biological weapons, then, come out of this comparison looking worse than other weapons. But it is not all bad news. Nuclear weapons are really hard to defend against; the current state of the art consists of threatening the other guy with a nuclear weapon of your own. With biological weapons, fortunately, we have better options.
For a start, we can keep them hard to build. Good enough technology might one day make a pandemic pathogen cheap and easy to produce, but we are not there yet; the first thing we can do is keep it that way, raising the bar to building one rather than letting it fall, for example by screening the synthetic DNA an engineered pathogen has to be assembled from. And if one is built and released anyway, the next thing is to catch it early, for instance by sequencing what shows up in clinics and in wastewater to flag a new pathogen before it spreads.
Second, for all the menace of self-replication, a pathogen still has to get inside you before it can do anything, and there are only so many ways in. It is also a living thing, and living things can be killed. Air filters, ultraviolet light, chemical disinfectants, and respirators could help defend us from even the worst possible threats.
Finally, because a pathogen has to replicate inside us, we can try to use drugs or other medical countermeasures to kill it before it kills us. This is, unfortunately, easier said than done; we still don’t have effective treatments for many natural diseases, and millions still die each year in poor countries due to diseases that are easily curable. But even if defensive technologies lag, there is still a lot we could do to be better prepared.
Appendix: Lethality estimates
All entries are order-of-magnitude estimates. The maximum deaths column assumes ideal conditions against a dense, undefended population. Costs are marginal procurement and operations for a well-supplied state, with reusable platforms amortized over their service life. R&D costs are likewise assumed amortized, with the exception of the pandemic pathogen, which can only be used once. The table covers only weapon systems that don’t require a human operator in the field. Hand-wielded weapons are cheap—a machete costs O($10) and can be used many times—but their lethality is limited by the operator, who is expensive and placed at high risk.
Pandemic pathogen: The maximum death count of one billion is a round figure for a highly transmissible, highly lethal agent that spreads globally. The cost, and its large uncertainties, are discussed in the main text. The estimate of 10⁻¹² g for a single infectious dose is conservative. Individual virions weigh 10⁻¹⁷–10⁻¹⁵ g, and infectious doses range from O(10) to well over 10⁶ virions.
Nuclear weapons: A standard 1 Mt thermonuclear warhead costs roughly $2.5M and weighs roughly 1,100 kg—the mass of the B83, the closest US example at 1.2 Mt. Delivered by airplane, a 1 Mt airburst over a dense city kills roughly 1.5 million (NUKEMAP); the amortized cost of the aircraft is negligible, giving a cost per death of roughly $2. A Trident D5 carries 8 W88 warheads, each 475 kt; spread across cities, the full payload could kill perhaps 4 million. The loaded missile costs roughly $80M and weighs 60 t.
Anthrax, by airplane: An aircraft spray of 100 kg of anthrax spores over a large city could kill 1–3 million (1993 OTA report); the table uses the midpoint of this range. The cost of producing and delivering the agent by light aircraft is roughly $200k (inflation-adjusted from the OTA-era estimate).
Anthrax, by Scud: A Scud carrying 30 kg of anthrax spores could kill 30,000–100,000 (OTA report); the table uses the midpoint of 50,000. A Scud-B costs roughly $1M (Zaloga) and weighs 5,900 kg, which dominates the total cost.
Conventional munitions, by airplane: A B-52 carries about 30 t of ordnance—roughly 130 Mk 82 (500 lb) bombs at $2,500 each. A 500 lb bomb has a lethal radius against unprotected personnel of roughly 50 m, for a lethal area of about 8,000 m² per bomb; 130 bombs cover a total lethal area of roughly 1 km². In a dense urban area (10,000 people per km²), this puts roughly 10,000 people in the nominal lethal area. Buildings provide substantial cover; as a rough upper bound, 1,000 deaths is plausible. For comparison, the Tokyo firebombing of March 1945 killed 100,000 with 1,665 tons of incendiaries at a density of 25,000/km²—about 60 deaths per ton—but incendiaries are far more destructive than high explosive against flammable cities like wartime Tokyo, which was largely built of wood. A 6-hour sortie costs $420k in flight time ($70k per flight hour) plus $300k in munitions, for a total of about $700k.
Conventional munitions, by Scud: The best data on conventional ballistic missile lethality comes from Lewis, Fetter, and Gronlund (1993), who use V-2 casualty records from London (518 impacts, 2,510 deaths) to build an empirical model. The V-2 warhead is comparable in size to the Scud-B (980 kg each). In London, the average death radius was only 21 m, giving a lethal area of 1,300 m² per impact. Scaled to the Scud-B’s somewhat larger energy release and a population density of 10,000/km², the model predicts roughly 20 deaths per strike—consistent with the Iran-Iraq War of the Cities, where 533 Iraqi missiles killed 2,312 Iranian civilians (about 4 per strike with sheltering). Scud costs of $1M and mass of 5,900 kg are the same as for anthrax.
Sarin nerve agent, by airplane: The OTA report and Lord Lyell (1996) estimate that a light aircraft dispersing 1,000 kg of sarin over a city kills 400–800—no better than the same aircraft dropping an equivalent weight of high-explosive bombs. A chemical sortie costs $200k.
Sarin nerve agent, by Scud: The same sources estimate 60–100 deaths from a Scud carrying 300 kg of sarin—modestly more than a conventional Scud warhead (20 deaths), but not by an order of magnitude. Chemical lethality is highly weather-dependent; these figures assume ideal conditions for the attacker.
Hobby drone with grenade: A consumer FPV drone dropping a grenade could kill maybe two people. Ukrainian frontline FPV drones cost roughly $400–500 per unit. The total mass is about 1.5 kg.
TIKAD gun drone: The TIKAD is an armed octocopter mounting a rifle or grenade launcher, killing roughly 10 per mission against undefended targets. No public unit price exists; the $6,000 per-mission figure assumes a $120k platform amortized over 20 missions, which is an order-of-magnitude estimate based on comparable systems. The drone weighs 2.5 kg.
QinetiQ MAARS ground robot: The MAARS is a remotely operated ground robot armed with a 7.62mm machine gun or 40mm grenade launcher, killing roughly 100 per mission with a full ammunition load. No public price exists; the $17,500 per-mission figure assumes a $350k platform (inferred from the comparable TALON/SWORDS at $230k) amortized over 20 missions. The robot weighs 8.5 kg.
MQ-9 Reaper: Armed with four Hellfire missiles ($115–150k each), a single sortie kills roughly 50. The marginal sortie cost is roughly $700,000—the four missiles plus operating costs—with the $30M airframe treated as sunk and amortized over its service life. The payload and fuel weigh roughly 1,400 kg.
Kinetic orbital bombardment: A 5-tonne tungsten rod delivers only tens of tonnes of TNT-equivalent on impact (discussion), so the 5,000 death toll requires a direct hit on a dense district. The launch alone costs roughly $5M at current commercial rates ($1,000/kg to orbit); the delivery platform adds roughly $10M, for a total of about $15M.
Some of the most dangerous non-physical entities also self-replicate. Computer worms spread between machines, dangerous ideas—from multilevel marketing schemes to genocidal ideologies—between minds, and a capable enough AI could copy itself across hardware. ↩︎
Nanotechnological “grey-goo” is an extreme example, though it would face resource constraints and vulnerabilities similar to those of biological organisms. ↩︎