An OpenAI model has just done something human codebreakers hadn’t managed in more than a century: ChatGPT‑6 Astra has fully solved a 108‑year‑old German World War I ADFGVX radio cipher.
The ADFGVX system
During WWI, the German military used the ADFGVX cipher for high‑value radio traffic. It’s based on a 6×6 substitution grid labeled with the letters A, D, F, G, V, X on both axes. Each pair like AA, AD, FX maps to one plaintext letter or digit. For example, the word “PRINZ” might be encoded as:
FX GD DX FV VD
Security comes from the keyword used to build and permute the grid. Many such messages have already been broken by cryptanalysts like George Lasry, but several remained on lists of unsolved ciphers.
One of them was a German radio message sent on November 27, 1918.
How Astra solved it
Astra treated the problem as a full ADFGVX reconstruction task and identified the key word “TRUPPENVERSCHIEBUNG” (“troop movement”), documented in J. Rives Childs’ The History and Principles of German Military Ciphers, 1914–1918 as a German key in late 1918.
The procedure:
- Alphabetically reorder the letters of the key
- Write the key as a header row
- Fill the ciphertext beneath it in columns
- Reorder columns according to the alphabetical key order
- Decode each ADFGVX pair via the reconstructed grid
The resulting German plaintext reads:
EIN ENGLISCHER KREUZER EINLIEG … SEWASTOPOL … EIN GESCHWADER DER ALLIIERTEN FOLGT 26STEN
In English: “An English cruiser arrived at Sevastopol on the 24th; an Allied squadron follows on the 26th.”
Checked against real ship logs
Astra then cross‑checked its own solution against historical sources and found that the British cruiser HMS Canterbury’s original logs show arrival at Sevastopol on 24 November 1918, with an Allied squadron arriving on 26 November.
Childs’ work suggests this key went into use on December 9, 1918, so this decode implies the keyword may have been deployed operationally earlier than previously documented.
For historians and cryptographers, the result shows how modern AI can assist not just in contemporary security research, but in re‑examining historical ciphers and military archives that have resisted manual analysis for decades.
Source: Prinz AI, Tom's Hardware
