When you think of wartime cryptography, you probably think of World War II and Enigma, although there were other famous codes used during that war. But in fact, there have been codes used through wars and in peacetime for much longer. Even the Romans used codes. Of course, World War I, or The Great War, as it would have been known, had its share of codes and, as you might expect, most of these have been broken long ago. Most of them. But apparently an AI model, GPT-6 Astra, recently cracked one that was previously undeciphered.
If you aren’t up on century-old cryptography, the ADFGVX cipher appeared in 1918. It began as ADFGX, but after a few months grew an extra letter — and a larger grid. The letters were chosen because their Morse-code patterns were relatively easy to distinguish: A, D, F, G, V, and X.
The original ADFGX version used a 5×5 grid containing the alphabet, usually merging I and J. The later ADFGVX version expanded this to 6×6, making room for all the letters and the digits as well.
Encoding required a 5×5 (or 6×6) grid with the alphabet within. The table’s rows and columns were labeled… you guessed it… ADFGX pr ADFGVX. Each plaintext letter gets replaced by the corresponding row and column such as AF or XG. Finally, the letter pairs would be written under a transposition key.
Concatenating those pairs gives AGXGDDDXXF. Written in five columns under CANDY, the first row is AGXGD and the second is DDXXF.
Next, you sort the letters in the transposition key, taking the columns along with the letters. That would make the transposition key ACDNY, and the first row would now be GAGXD. Then the message is sent by columns. Too much to wrap your head around? Visit dCode and plug in some plain text, squares, and keywords for yourself.
The French broke the code in 1918, but without computers it was tough work. Codebreaker [Georges Painvin] reportedly worked on the cipher until it made him ill, but he did solve it. However, some of the intercepted messages remained a mystery. One in particular was sent on November 27, 1918. Apparently, the message used the codeword TRUPPENVERSCHIEBUNG, a German key believed to have gone into service in December. If this decoding is correct, then this message used it even earlier.
The message was, of course, in German and appears to be a report about a British ship’s movements near Sevastopol. In fact, HMS Canterbury’s position on the dates mentioned matches the message.
Honestly, we were more impressed that people like [Georges Painvin] in 1918 were able to decode these intercepted messages. But it is still an interesting glimpse into the capabilities of Astra.
It seems that as long as there has been writing, there have been ways to send secret messages. The Swiss in The Great War had their own encryption method.
This is a lot more edifying than the “AI Cracked the Voynich” headlines we’ve been getting since 2018. Recurring stories claim a chatbot or mysterious neural network decoded the manuscript. A model can always output something, but nothing repeatable, coherent, and checkable against unseen pages has ever appeared. Pattern-matching is not reading.
Facts Only
* Romans used codes.
* The ADFGVX cipher was introduced in 1918.
* The original ADFGX version used a 5×5 grid; the ADFGVX version used a 6×6 grid.
* The cipher uses the letters A, D, F, G, V, and X.
* Georges Painvin decoded ADFGVX messages in 1918.
* A message sent on November 27, 1918, remained undeciphered until recently.
* The message used the codeword TRUPPENVERSCHIEBUNG.
* The message contains a report on British ship movements near Sevastopol.
* HMS Canterbury's position aligns with the details in the message.
* GPT-6 Astra decoded this specific message.
* The Swiss used their own encryption methods during World War I.
Executive Summary
Wartime cryptography has evolved from ancient Roman codes to the complex systems of the World Wars. During World War I, the German military utilized the ADFGVX cipher, a system combining a grid-based substitution with transposition. While French cryptanalyst Georges Painvin broke the cipher in 1918, certain intercepts remained mysterious for over a century.
Recently, the AI model GPT-6 Astra successfully deciphered a message from November 27, 1918. The decoded text describes the movements of the HMS Canterbury near Sevastopol, using the codeword TRUPPENVERSCHIEBUNG. This suggests the codeword was implemented earlier than previously believed. This achievement highlights the utility of modern AI in solving historical puzzles, contrasting with other high-profile but unverified claims regarding the decoding of the Voynich manuscript.
Full Take
The strongest version of this narrative is that AI is transitioning from speculative pattern-matching to verifiable utility by solving "cold cases" of cryptography. By providing a result that can be cross-referenced with historical naval records (HMS Canterbury), the claim moves from the realm of generative hallucination into empirical evidence.
The narrative operates on a paradigm of technological progression, contrasting the "brute force" intellectual labor of Georges Painvin with the algorithmic efficiency of GPT-6 Astra. There is an implicit assumption that the ability to crack a known cipher with a known key (or a discoverable one) constitutes a significant leap in "intelligence." However, there is a critical distinction being drawn between pattern-matching and actual reading—specifically regarding the Voynich manuscript. This suggests a growing demand for "repeatable, coherent, and checkable" outputs as the gold standard for AI validity.
The implication is a shift in how we value human agency in intellectual discovery. While Painvin’s struggle is framed as heroic and physically taxing, the AI's success is framed as an "interesting glimpse" into capability. The benefit here is the recovery of lost history, but the second-order consequence is the potential devaluation of the human cognitive process in favor of result-oriented computation.
Patterns detected: none
Root Cause: A transition from the "hype cycle" of AI (Voynich claims) to a "verification cycle" (historical cryptography).
Counterstrike Scan: An influence campaign would use this to claim AI possesses "superhuman" intuition or "consciousness" by attributing mystical properties to the decoding process. The actual content remains grounded in historical verification and does not match this pattern.
Bridge Questions:
1. Does the ability to decode a cipher with a finite key set prove linguistic understanding, or is it simply high-speed permutation?
2. How does the verification of a historical fact change our trust in AI's claims regarding texts where no "ground truth" exists?