The swarm
that is coming

Start with the one thing you have to understand. Everything else comes out of it.

A cartel drone today works like this: a man hides a kilometer away, watches a screen, flies the machine with his hands, drops the load and leaves. He is a sniper with wings. Dangerous, yes. But one at a time, against one target.

A swarm works differently, and the difference is not quantity. It is nature. In a swarm nobody flies anything. The operator marks a zone and the software divides the work among a hundred machines that talk to each other, correct themselves, replace whichever one falls and close the formation again. Shoot down twenty and the remaining eighty recalculate and keep going.

It is the difference between a man with a rifle and an infantry company. Except the company does not eat, does not sleep, does not desert and has no mother to claim it.

Why this breaks Mexico's defense

The Mexican state has answered violence with one formula for twenty years: send more men. More soldiers, more national guard, more convoys. That formula works against an enemy who fires bullets.

Against a swarm it inverts. Concentrating troops stops being strength and becomes a target. A convoy is exactly what a swarm is designed to destroy: many objectives bunched together, no roof, no dispersion. Same for a checkpoint, a base, a rally, a polling station, a refinery.

And the arithmetic of money inverts too. A commercial machine costs what a motorcycle costs. Defending a single building against a swarm costs what a helicopter costs. The attacker spends little and the defender spends everything. That asymmetry is what turned the Houthis, a group from a ruined country, into an actor capable of shutting down Red Sea commerce.

So this investigation is not about drones. It is about the date the Mexican government loses the advantage of having more men.

Here is the number nobody has put on the table, and it is the heart of all of it.

The cartel does not need more people to have a swarm. It needs the same handful of operators it already has, with different software on top. What multiplies is not the payroll. It is what that payroll can reach in one night.

Multiply that by the size of the unit. We documented that CJNG already has one, with patches and its own training, and that in 2023 it taught another cartel to do the same.

Ten operators who today can hit ten points in one night hit a thousand with a swarm. It is not a bigger force. It is a different country.

And there is a political consequence almost nobody has said out loud. With that capability, a cartel stops negotiating territory with the government and starts negotiating another way: it can shut down an airport, paralyze a refinery, make an election impossible in an entire state. It stops being a security problem and becomes a sovereignty problem.

Nothing that follows is interpretation. Every line has a primary source behind it, and it is cited.

These are the five. Each one is measured from zero to ten, where zero is having none of it and ten is having all of it. This is how CJNG scored.

Read them top to bottom and you see the problem at a glance: on four of the five the cartel is well supplied. The short bar is the one that decides.

The four who already went through this

Two crossed fast, one did not cross despite having everything, and one crossed with a state helping. The rule comes out of that.

This is where the model says something nobody wants to hear, which is why it is worth reading slowly.

Everyone assumes two things. That the more violent a group is, the closer it gets to bigger weapons. And that the more money it has, the sooner it gets them. Both are false, and the four precedents prove it.

Hezbollah had everything. Money, its own engineers, and Iran handing it technology for twenty years. It did not cross to swarms. What it lacked was hurry: it was not fighting for its existence.

ISIS had almost nothing. No state backing it, no reserves. It crossed in under two years. It had two things: engineers and smuggling.

That is where we drew the rule from, and it fits in one line. You need urgency, plus one of two doors: either a state that supplies you, or your own people who know how.

Now run CJNG through that rule, with the numbers we calibrated.

Money it has to spare, 0.70. Smuggling, which is its trade, 0.80. Urgency, since El Mencho was killed, very high. A state supplying it, almost none, 0.20.

It has everything but one thing. Its own people who know how: 0.50. And in this arithmetic a chain is worth what its weakest link is worth. That is why the final index comes out at 0.50 and no higher.

The test that changes everything

We fed the model what happened in February. Twenty-five national guards killed, reprisals across twenty states, urgency spiking to 0.90.

The index did not move. Still 0.50.

Then we raised technical talent by a single tenth, from 0.50 to 0.60. The index jumped to 0.60 and CJNG landed on the same side as ISIS and the Houthis, the two that did cross.

Read it again. All the blood of February did not bring the cartel one day closer to the swarm. One hired engineer does.

And here is the good news, which is also the hard part

If the bottleneck were money, Mexico would be lost: the cartel has more than the state. If it were violence, lost too: it has all of it.

But the bottleneck is people. And the people who can build this in Mexico are not millions, or thousands. They are engineers in electronics, computer vision and flight control. In the whole country they fit in one auditorium. Maybe fifty of them decide the matter.

Which means this problem, unlike almost every other one in Mexico, can actually be solved, and costs less than any military operation. You do not have to win a war. You have to get a few dozen people to take a different job, and get the ones already inside to walk out.

Every one who joins moves the clock forward. Every one who leaves, by arrest, by defection or because someone offered them something better, takes it with them.

The count that matters

The paper was presented in February 2026. Twelve months from then is February 2027. Eighteen is August 2027.

If state pressure eases, the clock moves up six months. If pressure rises on talent and inputs at the same time, the hybrid path slides to a range of eighteen to thirty months. The decision window is not in 2027. It is now.

We documented a semi-formal unit inside CJNG. They are called the Operadores Droneros. They have patches, they have training, and in 2023 they transferred capability to La Familia Michoacana.

That unit is not an anecdote. That unit is the INTCAP condition. And INTCAP is the number that binds the whole system.

Translated into public policy, with no academia in the way: every engineer, every technician, every ex-soldier who joins that unit is worth more to the swarm than any highway ambush. And every one who leaves, by arrest, defection or cooperation, stretches the clock for everyone.

The three countermeasures we propose for the internal path attack exactly that variable and no other: infiltrate recruitment, neutralize workshops, and control the export of computing and vision modules.

The second early warning

The other door, EXTSUP, is nearly shut at 0.20. But we flagged something worth underlining: Chinese and Iranian drone supply channels already fed Russia and the Houthis. Those pipes exist and they work.

A single state-backed supplier opens that door. That is the second indicator to watch, and the one that would change the clock overnight.

Here is the hook

A foreign terrorist organization led by a United States citizen needs no extradition to be tried under United States counterterrorism law.

And the swarm, by definition, gets built out of what we call modules in the model: computing, vision, navigation. Every module that crosses the border is a document. Every document is a jurisdiction.

The missing document

To move this from academic inference to case file, one thing is needed, and it is already identified: the Bureau of Industry and Security license and denial record for exports to Mexico of embedded computing modules, computer vision and RF links classified under ECCN 4A and 5A, from 2023 to 2026, cross-referenced against the SDN list.

FOIA request to BIS: in preparation. That document turns a projection into a file.

What could not be proven

We say it before anyone else does. There is no direct public evidence that CJNG has bought artificial intelligence modules for drones. We infer it from its purchasing practices in other sectors and from the compared cases. There is no official drone incident figure for 2026. The Michoacán prosecutor file on Flores Silva could not be obtained. And the calibration of the five conditions is expert judgment over open sources, robust to plus or minus 0.33 by our own testing, not direct measurement.

Right of reply is open. Any response from the people named is published in full on this page.