In the peculiar annals of interwar intelligence, few episodes better illustrate the intersection of human frailty and state security than the evolution of Soviet encryption between the world wars. The story involves poetry, vodka, and one of cryptography’s most colorful characters – Ernst Fetterlein, the former Czarist chief cryptanalyst who walked across the Finnish border during the Revolution to join British intelligence.

The Soviets’ cryptographic journey from 1920-1928 showcases a remarkable evolution. They began with surprisingly basic columnar transposition[1] of Russian plaintext, moved to dinomic substitution[2] before transposition[3] in late 1920, then progressed through increasingly complex systems. One of these systems fell to British cryptanalysts from a simple literary connection. When Fetterlein couldn’t figure out the keys using Russian letters, Tiltman discovered they were actually from an obscure out-of-print pocket edition of poems by George Wither, a prolific 17th-century English poet. Tiltman noted:

I do not remember the method of indicating keys, but I know it was simple and that, after finding the source book, we were in a position to decrypt DELEGAT messages as soon as the intercepts reached us.

By 1928, the Soviets finally adopted one-time pads (OTP), using two specific formats where messages over 1100 figures had to be split into parts. The pads were used “boustrophedon” style (from ancient Greek, meaning like an ox turning while plowing), and critically, operators were officially permitted to use each pad twice but no more. This policy had been built into their procedures from the beginning in 1928, not as a wartime compromise as long believed:

275-figure pad (11 lines × 5 groups of 5 figures) for messages up to 550 figures:

12345 12345 12345 12345 12345  →
54321 54321 54321 54321 54321  ←
12345 12345 12345 12345 12345  →
54321 54321 54321 54321 54321  ←
12345 12345 12345 12345 12345  →
54321 54321 54321 54321 54321  ←
12345 12345 12345 12345 12345  →
54321 54321 54321 54321 54321  ←
12345 12345 12345 12345 12345  →
54321 54321 54321 54321 54321  ←
12345 12345 12345 12345 12345  →

550-figure pad (11 lines × 10 groups of 5 figures) for longer messages:

12345 12345 12345 12345 12345 12345 12345 12345 12345 12345  →
54321 54321 54321 54321 54321 54321 54321 54321 54321 54321  ←
12345 12345 12345 12345 12345 12345 12345 12345 12345 12345  →
54321 54321 54321 54321 54321 54321 54321 54321 54321 54321  ←
12345 12345 12345 12345 12345 12345 12345 12345 12345 12345  →
54321 54321 54321 54321 54321 54321 54321 54321 54321 54321  ←
12345 12345 12345 12345 12345 12345 12345 12345 12345 12345  →
54321 54321 54321 54321 54321 54321 54321 54321 54321 54321  ←
12345 12345 12345 12345 12345 12345 12345 12345 12345 12345  →
54321 54321 54321 54321 54321 54321 54321 54321 54321 54321  ←
12345 12345 12345 12345 12345 12345 12345 12345 12345 12345  →

This brings us to the rather spectacular case of cipher clerks Kotlov and Serafimowich in 1926 Kabul. Serafimowich’s frequent hangovers led to so many encryption errors that Moscow mandated all messages be signed by the encrypting clerk. This well-intentioned administrative solution spectacularly backfired – the signatures provided British analysts with precisely the known-plaintext material they needed. When Serafimowich himself later decrypted orders for his own recall to Moscow over questionable papers, he fled to the British Embassy but was turned away. As Brigadier Tiltman notes in his recently declassified papers, he was “never heard of again.”

Yet even with these procedural vulnerabilities, Tiltman’s documents reveal something surprising – British cryptanalysts “were hardly able to read anything at all except in the case of one or two very stereotyped proforma messages.” The theoretical strength of the one-time pad still made decryption extremely difficult, even when used twice.

The ramifications would echo through cryptographic history. Operation Venona’s later success in breaking these reused pads became one of cryptanalysis’s first major computer applications, ultimately exposing Soviet agents like Blunt and Cairncross.

Over 100 Security Service (MI5) files are being released today [14 January 2025] covering a wide range of subjects and individuals. Most notably, the files offer fresh perspectives on notorious members of the Cambridge Five spy ring, namely Anthony Blunt, Kim Philby and John Cairncross.

The “two-time pad” problem may have helped shape modern information theory itself. Claude Shannon’s groundbreaking post-war papers establishing the mathematical foundations of cryptography likely drew inspiration from the practical challenge of quantifying exactly why a one-time pad was secure but a two-time pad catastrophically weak.

The serious implications for modern national security systems remain stark: mathematical perfection means little without robust operational security. The Soviets’ experience shows how administrative decisions and human reliability issues can compromise even theoretically unbreakable systems. Their requirement for cipher clerks to sign their work, like artists initialing a canvas, transformed a personnel management solution into a critical cryptographic vulnerability.

As we consider contemporary challenges in vetting national security roles, this history serves as a pointed reminder: institutional failures to properly handle personnel reliability issues can cascade into catastrophic failures. Though the specific technologies have changed, the fundamental challenge remains: a security system is only as strong as its human operators and the procedures governing them.

The Soviet Union’s cryptographic evolution through the 1920s demonstrates a typical progression from basic to multi-layered:

[1] Columnar Transposition
A message is written in rows of fixed length to form columns, which are then rearranged using a numeric key. For example, with key “3142” the message “SEND SUPPLIES” becomes:

  3 1 4 2
  S E N D
  S U P P
  L I E S

Reading columns by key order (1,2,3,4) produces: “EUI DPS SSL NPE”

[2] Dinomic Substitution
Plaintext letters are first paired (e.g., SE ND SU PP LI ES) and each pair is then substituted as a unit using a conversion table. This approach conceals single-letter frequency patterns that plague simple substitution. For instance:

- SE → KR
- ND → MY
- etc.

[3] Combined Method
This advanced technique applies both methods sequentially:

1. Applying dinomic substitution
2. Performing columnar transposition

The combination strengthens the encryption in two ways:

• Substitution masks letter frequencies
• Transposition scrambles positional patterns

This progression through increasingly sophisticated encryption methods reveals how deeply British signals intelligence had penetrated Soviet-Afghan diplomatic communications. The radio intercept stations at Cherat (above Peshawar) and Pishin in Baluchistan were particularly focused on monitoring traffic between Moscow-Kabul and Moscow-Tashkent, providing significant advantages in managing regional power dynamics and potentially influencing events covertly (e.g. King Amanullah Khan’s overthrow).

In the fall of 1925 the Government of India sent a column (known as the WANA column) to the northwest frontier to occupy Waziristan to deal with unrest among the northwest tribes, a more stormy situation than usual. Stark, the Russian Ambassador in Afghanistan, sent a cipher telegram to Moscow in which he inquired what joint action was proposed between the Russian and Afghan Governments “in view of the occupation of Waziristan (W Widu Okkupacii Waziristana).” Our interpreter, who was quadriligual in Russian, English, French and German, but not outstandingly literate in any one of them, translated this—”with a view to the occupation of Waziristan.” The intelligence branch of Army Headquarters was in Delhi, and we were in Simla, and there was a day of near crisis in Delhi before someone, realizing that it would take something like six months for Russians and Afghans to join forces over the Hindu Kush, queried the translation back to us.

The WANA column incident perfectly illustrates how signals intelligence capabilities needed to be balanced with practical regional knowledge. While the British intercept stations could successfully capture and decrypt Soviet diplomatic traffic, the true value of this intelligence depended on accurate translation and interpretation by cooler heads. In this case, a simple preposition mistranslation nearly triggered a crisis until those familiar with the Hindu Kush’s geography could provide crucial context about the practical impossibility of rapid Soviet-Afghan military coordination.

The recent Foreign Affairs piece on American authoritarianism fundamentally misses how AI will supercharge authoritarian power in unprecedented ways.

The Path to American Authoritarianism
What Comes After Democratic Breakdown
Steven Levitsky and Lucan A. Way
February 11, 2025

While the authors correctly identify the risk of democratic breakdown, their analysis is unfortunately trapped in an outdated framework that fails to grasp two critical accelerants.

First, they underestimate how AI already weaponizes America’s buried atrocities. Unlike human narratives that often gloss over historical trauma, AI can instantaneously surface and connect centuries of state violence by normalizing it – from President Jackson’s genocidal Trail of Tears to President Wilson’s Red Summer of 1919 leading to Tulsa Massacre of 1921. AI doesn’t miss the subtext of racist deception in the Missouri Compromise or the brutally racist and illegal conquest of Texas and Florida to expand slavery. It can relentlessly illuminate how “America First” movements always consistently and repeatedly enabled American race-based authoritarianism since the late 1800s.

The authors vaguely suggest institutional guardrails could contain authoritarian power. But they fail to recognize how AI can weave foundational historical threads into devastating narratives that undermine faith in those very institutions. When AI connects the dots between a past of systemic state violence and the present institutional power of non-governmental “efficiency” (totalitarian) mercenaries called DOGE, it becomes much harder to believe in the protective power of courts or federalism.

Second, they dramatically underestimate the velocity of AI-powered narrative control. Their analysis feels like watching someone explain how decades of prior print media will hold the line on public opinion in 1933, while completely missing how the Nazi regime flooded radio waves and totally rewrote reality in just three months. 2025 AI is far more powerful than 1933 radio – it can generate, target, and amplify hateful messages at a scale that makes Hitler’s genocidal machines look primitive.

The authors worry about gradual institutional capture through bureaucratic maneuvering. But they miss how AI can simply flatten institutional resistance through overwhelming narrative force.

Why bother carefully pressuring judges and abiding by them when AI can flood every platform with unaccountable sock puppet messages demanding targeting judges to be eliminated if “woke” or opposed to “efficiency”? Already the White House has announced they will be “looking into” any judge who disagrees with “efficiency”. The speed and scale of AI-powered propaganda makes the old ways of careful institutional analysis feel quaint, like marching troops with slow-firing inaccurate muskets into a machine gun. Domain shifts are devastating to analysis that doesn’t account for what’s changed.

Therefore the Foreign Affairs assessment is not just wrong, it’s dangerously overconfident in the way that reduces opposition to mass unjust incarceration and death. By suggesting American institutions can weather authoritarian pressure through quaint concepts of traditional resistance, they underestimate how AI already fundamentally changes the game.

Quantum threats are basically here and some people still don’t know how to change their passwords.

Does anyone really think executive orders pumped out by the hundreds aren’t being written with software? Does anyone really not understand why a few college-aged kids who barely write software are being called “auditors” of “efficiency” on a highly complex financial system they can’t possibly understand?

They are feeding all, and I mean all, American citizen data into Elon Musk’s private unsafe AI infrastructure and asking it “what would Hitler do, in the voice of Goebbels?”

This won’t be a slow erosion of democracy through bureaucratic weaponization, waters creeping up on those who don’t have boats. It already is a tsunami-level warning of AI-powered narrative control that will catastrophically sweep away democratic institutions faster than any previous authoritarian transition.

The authors claim America won’t face “classic dictatorship.” But by failing to grasp how AI supercharges authoritarian power, they miss that we’re facing something potentially worse – a form of technologically-enhanced authoritarianism that could exceed anything in history. And this “Technocracy” disaster has been many decades in the making, a Musk family obsession since the 1930s as proven out in South African apartheid, not something political scientists today should be unfamiliar with.

In case Elon Musk’s encoded speech pattern is unclear, planet “Mars” is used (incorrectly) to promote open violation of the law and disobeying law enforcement, like saying America will finally be as good as Mars when the white men who occupy it can’t be regulated: Occupy Mars = Aryan Nation.

We all know the children’s tale about what comes next if we don’t understand the threat. The institutional safeguards appear as straw huts against a coming huff-a-puff wolf. We need to wake up to the true scale and speed of the threat before it’s too late.

The simple reality is this: AI-powered authoritarianism won’t respect and carefully navigate around slow democratic institutions – it will overwhelm them in raw narrative force at unprecedented speeds causing disasters to force surrender and complacency.

To put it another way, Nazi generals carelessly sped full speed into France to overwhelm their targets while leaving themselves dangerously exposed. The French capitulated and resolved themselves to occupation instead of rapid counter attack that would have destroyed the Nazis. General Grant understood this in the 1860s yet the French didn’t grasp adequately the domain shift tactics of radios, planes and trucks.

The Foreign Affairs authors are analyzing how to defend against 20th century authoritarianism while missing that an AI invasion force already has landed and is expanding. They’re not just wrong about defenses, they’re complacent and leaving America dangerously unguarded.