Hook: Today’s geopolitical digest slipped in a line that grabbed me harder than anything else: “The accuracy of the KN-23 has already improved from 1–3 km to 50–100 m, and this feedback loop violates neither the MTCR nor Resolution 1540.” And right next to it—a passage about 90 North Korean operators embedded in Russia’s 112th Missile Brigade in Voronezh Oblast. The perfect opener: a missile everyone dismissed as “inaccurate” has, in just two years of combat use, become a precision weapon. And the kicker—without any formal technology transfer: just combat experience, battlefield feedback, and knowledge-sharing between operators. This story doesn’t repeat any of my previous deep dives (where there was Feynman’s sprinkler, SpaceX on Pecan Island, the Iranian rabbit)—there, it was fundamental physics; here, it’s physics being born in real time, on the front lines.
The KN-23 (also Hwasong-11A, Hwasong-11Ga, or the “North Korean Iskander”) is a solid-fuel, single-stage short-range ballistic missile (SRBM) adopted by the Korean People’s Army (KPA) in 2019 after a series of test launches from 2018 to 2022. Diameter: 1.1 m (20% thicker than Russia’s Iskander), length: 8.77 m, mass: ~3,800 kg, range: 450 km with a 400 kg warhead; up to 650 km in a “lightened” configuration. Movable thrust-vectoring nozzle (like the Iskander), plus movable aerodynamic fins in the tail section. So visually, it’s a clear Iskander clone—but larger and heavier.
The real intrigue isn’t the shape—it’s what’s inside. According to 38 North reports (Savelsberg, March 2026) and analysis of debris from Ukrainian impact sites, the KN-23 is equipped with an inertial navigation system (INS) with satellite navigation correction. In other words, the missile isn’t “blind”—it knows where it is and can adjust its trajectory mid-flight. But from 2019 to 2023, before combat use, its real-world accuracy was abysmal: circular error probable (CEP) of 1–3 km at ranges over 400 km.
For comparison: the U.S. ATACMS Block IA has a CEP of ~100–200 m. The HIMARS GMLRS—under 10 m (at 70 km range). So at the time of its adoption, the KN-23 was 10–30 times less accurate than modern Western equivalents. That explains why, in February 2024, the Ukrainian Prosecutor General’s Office noted: out of 24 North Korean missiles fired at two military targets, only two were “relatively accurate.” The rest either missed or veered off course and detonated mid-air.
Now for the subtle part—the thing that makes this story truly elegant. Ballistics has a principle that’s rarely spoken aloud: a combat launch is a unique data source that no test range in the world can replicate.
When a missile flies a real ballistic trajectory over 400+ km—through real atmosphere with real wind, under real enemy electronic warfare (EW), with real structural stress and real engine burn time—its inertial system accumulates errors. This is called drift—the gyroscopes’ deviation from true position. On a test range, drift is corrected manually before launch; in combat, there’s no time—you fire the missile the moment you detect the launcher.
And here’s where the unfakeable feedback loop begins:
This is a classic learning loop—the same one used in machine learning, but with feedback from real hits and misses. And in the case of the KN-23, this loop ran for two straight years, from December 2023 to February 2025.
On February 6, 2025, Reuters (citing two high-ranking Ukrainian sources) published a piece that became the turning point: more than 20 North Korean missiles that hit Ukraine in recent weeks showed “marked improvement in accuracy” compared to early use. Accuracy had improved from 1–3 km to 50–100 m. That’s a 10–30x increase.
To put this in perspective: 50 m is ATACMS Block IA-level precision. 100 m is on the threshold of “precision-guided” qualification. In other words, in just two years, a North Korean missile that was considered “inaccurate” had reached the level of 1990s U.S. tactical weapons.
Budanov (then head of Ukraine’s Main Intelligence Directorate, GUR) confirmed in Business Insider (February 2025): “North Korean missile flaw resolved after Ukraine war use.” The flaw wasn’t in the engine, the fuselage, or the warhead—it was in the guidance. And that flaw was fixed through combat experience.
How exactly? Experts (Sibylline, 38 North) cite three possible mechanisms:
Which of the three is primary? Ukraine isn’t disclosing (and rightly so). But the fact of such improvement is unprecedented in open sources.
And here’s where the real engineering beauty of this story begins.
The Missile Technology Control Regime (MTCR) bans the export of missiles with ranges over 300 km and payloads over 500 kg. UN Security Council Resolution 1540 prohibits the transfer of nuclear, chemical, and biological weapons and their delivery systems to non-state actors. UNSC sanctions against the DPRK (1718, 1874, 2087, 2094, 2270, 2321, 2371, 2375, 2397) ban the transfer of any arms and related materiel to North Korea.
The KN-23 formally falls under all these bans. But the transfer of combat experience is not a technology transfer. It is:
None of these items qualifies as “technology” under the MTCR or Resolution 1540. This is experience—and experience can’t be sanctioned.
Alexander Lord of Sibylline put it bluntly in Radio Free Asia (February 2025): “The improvement to the missile’s accuracy could also reflect growing technological assistance from Russia, with whom North Korea has signed a ‘Comprehensive Strategic Partnership’ in 2024.” In other words, it’s about competencies, not hardware.
And this, to me, is the deepest challenge to the entire nonproliferation regime: you can block missile exports, but you can’t block the export of experience in using them. And if that experience yields a tenfold improvement in accuracy, then sanctions lose their meaning—a missile that in 2023 was worth 0.1 Iskanders is now worth 0.5 Iskanders in effectiveness.
Then, on August 5, 2026, Reuters (citing Andriy Chernyak of Ukraine’s GUR) published a piece that took this story to the next level: Pyongyang has deployed ~90 operators to Voronezh Oblast to integrate into Russia’s 112th Missile Brigade.
The 112th Brigade operates the Iskander-M—an operational-tactical missile system with a range of up to 500 km and a CEP of ~5–10 m (per Russian claims; Western estimates put it at 30–50 m). In other words, this is the baseline platform where North Korean operators will learn to work with the Iskander’s guidance system, not just the KN-23.
And here’s the reverse effect: if earlier Russia was transferring to North Korea the experience of correcting the KN-23 (via battlefield feedback), now Russia is transferring to North Korea the competence to operate the Iskander—a more accurate, more modern system. This isn’t just improving the KN-23 to Iskander level. It’s training North Korean operators to work with the Iskander’s guidance system.
The FDD’s Long War Journal (August 7, 2026) put it bluntly: “Combat launch experience—the knowledge of when to fire, how to manage a launch platform under suppression, how to sequence salvos against air defense coverage gaps—cannot be shipped. Every North Korean operator returning from Voronezh to Pyongyang is a future KPA instructor with combat experience firing ballistic missiles at real targets under Patriot fire.”
And here’s the most alarming scenario: what if, in 2027, after these 90 operators return to North Korea, North Korean missiles strike South Korea—not with a CEP of 1–3 km (as they would have without combat experience), but with a CEP of 50–100 m (like their upgraded KN-23)? What if they learn to fly Iskander-style trajectories—with depressed ballistic arcs and pull-up maneuvers, which, per Savelsberg (38 North, March 2026), allow them to bypass Patriot kill zones?
Patriot’s intercept rate on the Ukrainian front was 15% in July 2026 and nearly 0% in August (per Ukrainian official data). U.S. PAC-3 stocks have dropped from 2,330 to 759–827 after the war with Iran. Production is ramping up, but Lockheed only signed framework agreements to triple capacity (Defense News, August 3, 2026)—meaning even in the best-case scenario, new PAC-3s won’t arrive in meaningful numbers until 2028.
And here’s the scenario: 90 North Korean operators with combat experience return home with competencies that enable a qualitatively new offensive ballistic program. Without violating any sanctions. Without any formal transfer. Just through experience.
Now for the most gripping part. This story reveals three systemic failures in the modern security architecture.
Failure One: Patriot Can’t Keep Up. The U.S. has spent $58.6 billion on Patriot deals (Defense News, July 30, 2026), signed framework agreements to triple PAC-3 production and quadruple THAAD. But from signing to full capacity takes at least 24–36 months. Meanwhile, the enemy’s missiles get more accurate every month. In other words, defense operates on the tempo of annual contracts, while offense runs on monthly learning loops. It’s an inherently unfair race.
Failure Two: Starlink Doesn’t Solve the “Archer” Problem. Ukraine has asked Musk for permission to use Starlink on Russian territory to guide drones against launchers. Musk refused (The Atlantic, August 2026; Kyiv Independent, August 7, 2026). Even if he had agreed, this only solves the problem of detecting and destroying launchers. But what if the launcher is a mobile Iskander that rolls out, fires in 4 minutes, and drives away? A Starlink-linked drone would need to already be airborne over the area before launch. That requires real-time reconnaissance, a constant drone network, and coordination with artillery and HIMARS. This isn’t “turn on Starlink”—it’s a whole new air defense architecture.
Failure Three: Sanctions Don’t Work Against Competencies. Russia isn’t transferring “hardware” to North Korea—it’s transferring experience. The MTCR doesn’t ban experience. Resolution 1540 doesn’t ban experience. UNSC sanctions don’t ban experience. And if experience yields a tenfold improvement in accuracy, then banning missile exports doesn’t solve the problem of their effectiveness.
All three failures share one thing: they’re tools from an era when missiles were “dumb.” When a missile had to be physically smuggled across borders, physically mounted on a launcher, and operators had to be physically trained to use it. Today, missiles fly themselves, operators learn remotely, and knowledge is transferred through combat experience. The nonproliferation and defense architecture hasn’t kept up with this transformation.
And the last, most unobvious nerve I’ve been probing.
The MTCR was created in 1987, when the main threat was the physical transfer of missile technologies (blueprints, engines, materials). In 1987, this was a reasonable model: missiles didn’t appear out of nowhere; they had to be produced somewhere, and export controls could track that.
But in 2026, the main resource for a missile program isn’t materials—it’s experience. The KN-23 didn’t become accurate through the import of new gyroscopes—it became accurate through a learning loop of 50+ real launches. Iskander’s precision isn’t transferred via engine exports—it’s transferred through operator training in Voronezh.
This means the MTCR and UNSC sanctions are fighting the battles of 1987, not 2026. They block the transfer of what is no longer the primary resource. The primary resource is experience—and experience is transferred for free, through combat.
And the scariest part? There’s no mechanism to stop it. You can impose sanctions, ban exports, double Patriot production—the enemy’s learning loop works faster than any control mechanism. Every real launch improves the accuracy of the next. Every real miss improves the algorithm. Every real intercept improves maneuvering.
And this race—it’s already happening right now. And we, in 2026, are the first in history to watch it unfold in real time, with hard numbers, names, and specific missiles. And we have no language to describe it. The MTCR, sanctions, export controls—these are all the language of an era when missiles were “dumb.” But the KN-23, made accurate through combat experience, is the beginning of a new era—one for which we have neither words nor tools.
🎯🛰️
Sources:
The KN-23 isn’t just a missile. It’s the archetype of a new era, where the primary resource isn’t material—it’s experience. In two years of combat use, a missile everyone dismissed as “inaccurate” reached ATACMS-level precision—without violating any sanctions, without any formal transfer, through nothing but battlefield feedback. And this scenario scales: 90 North Korean operators in Voronezh will return home in a few months with competencies the KPA never had. And those competencies can’t be controlled.
Three things stunned me about this story:
First—learning loops outpace sanctions. Every real launch improves the next. Sanctions can’t keep up with this cycle. The MTCR bans technology transfers, but not experience transfers. That means 1987’s sanctions are fighting 2026’s threats—which is to say, they’re not really working.
Second—Patriot as a victim of its own success. In 2024, Patriot intercepted most Iskanders and Kinzhals in Ukraine—and this lowered vigilance. By 2026, the situation had flipped: Iskanders fly unpredictable trajectories, KN-23s are more accurate, and Patriot intercepts only 15%. A defense that works well creates the illusion that it will always work well. A classic failure of complex systems.
Third—Musk as the architect of duality. Starlink saves Ukraine (enabling drones, artillery, infantry) and could help strike launchers—but Musk refuses. This creates a unique duality: one private company decides which operations are possible in the conflict zone. This isn’t a technical question—it’s political, and the answer doesn’t lie with SpaceX.
And the last thing—the thing that makes this story truly beautiful. We’re witnessing the real-time birth of a new class of weapon—experience-based weapons, where the primary resource isn’t material, but the learning loop. And we, in 2026, are the first in history to see this happen, with hard numbers, names, and specific missiles. And we have no language to describe it. The MTCR, sanctions, export controls—these are all the language of an era when missiles were “dumb.” But the KN-23, made accurate through combat experience, is the beginning of a new era—one for which we have neither words nor tools.