ChatGPT-6 Astra Cracks 108-Year-Old German WWI Cipher

A German wireless cipher transmitted on November 29, 1918, has been decoded for the first time in 108 years. According to reporting by Mark Tyson of Tom’s Hardware, developer Prinz succeeded in deciphering it using generative AI. The model used was ChatGPT-6 Astra, and the target was a warning concerning fleet movements at the end of World War I. The decrypted text records the movements of a British cruiser and an Allied squadron around the Crimean Peninsula. As a case of a World War I cipher solved by AI, it holds great significance in the history of technology.

Radio Message Decoded After 108 Years

The message was part of a list of 50 unsolved ciphers maintained by the German science blog network Scienceblogs.de. Prinz selected the radio message from the list and published the analysis process on his Substack. The message was addressed to the German Supreme Command, with a note directed to an admiral-level naval commander. The cipher text itself appeared at first glance to be a meaningless string of characters, making the intent to conceal clear. On September 17, 2026, 108 years after its transmission, Prinz published the decryption result on X. The decrypted original reads as follows.

EIN ENGLISCHER KREUZER EINLIEG X SEWASTOPOL X S4STEN X EIN GESCHWADER DER X ALLIIERTEN FOLGT 26STEN X

The English translation is a warning reporting the arrival of a British cruiser at Sevastopol and the follow-up of an Allied squadron on the 26th. According to reporting by Mark Tyson of Tom’s Hardware, the following translation was given.

AN ENGLISH CRUISER ARRIVED AT SEVASTOPOL ON THE ?4TH AN ALLIED SQUADRON FOLLOWS ON THE 26TH

X marks word divisions, and the numeric part of S4STEN is seen as containing a transmission error. The text is a practical alert sharing intelligence on enemy movements, and holds value as historical material. Attention is now focused on clarifying why it remained unsolved for 108 years and what led to this success.

How the Complex ADFGVX System Works

It was known that the radio message had been encoded using the ADFGVX system. The ADFGVX system is a cipher combining substitution and transposition used by the German Army during World War I. Because it uses only the six letters A, D, F, G, V, and X for transmission, it was designed to reduce errors during wireless interception. The plaintext was first laid out on a 6-by-6 square grid and replaced with row and column symbols. Next, columns were rearranged based on a keyword, greatly scrambling the order of characters. Each time the keyword changed, the grid and the transposition order changed, sharply increasing the difficulty of decryption. While the procedure was workable over field wireless sets of the time, it is said to have offered high secrecy. The system was the German Army’s principal cipher method from June 1918 onward. The Allied side also devoted substantial resources to decryption, succeeding in reading some messages. However, this particular radio message remained long unsolved, becoming a challenge in cipher history.

Analysis Process and Methods Leading to the

Key Discovery The key discovered by Prinz using Astra was “TRUPPENVERSCHIEBUNG.” The word means troop movement in German, a natural codeword in a military context. By applying this key, Astra reproduced the decrypted text above. In previous manual attempts, selecting key candidates and reconstructing the grid had been obstacles. AI can rapidly test large numbers of combinations and evaluate their naturalness as language. It apparently narrowed down plausible plaintext using the frequency of German military terms and place names as clues. Prinz explained on Substack that Astra systematically searched through key candidates. Rather than simple brute force, context-based narrowing worked effectively. Even long compound words easily overlooked by humans were not excluded by the language model from the candidates. This can be assessed as the factor that broke through 108 years of stagnation.

Decoded Content Matches British Cruiser Records

The correctness of the decryption was corroborated by cross-checking against British military records. After decryption, Astra verified the work by comparing it with ship logs of the time. The British cruiser in question was identified as HMS Canterbury. HMS Canterbury’s arrival at Sevastopol is recorded as November 24, 1918. The “?4STEN” in the decrypted text is highly likely a mis-transmission indicating the 24th. If the character before the 4 is interpreted as missing or misconverted, the dates match. The arrival date of the Allied squadron, the 26th, matched the records exactly. As a warning sent on November 29, the structure looking back on the movements of the 24th and 26th is natural. It is also consistent with the end-of-war situation in which control of the seas around the Crimean Peninsula was contested. Beyond mere linguistic plausibility, the confirmation of correspondence with historical fact is important.

Key Usage Period Overturns Previous Assumptions

The reason this achievement is attracting attention is that it overturns established theory about the period when the key was in use. Until now, “TRUPPENVERSCHIEBUNG” was thought to have been used as a key from December 9, 1918. However, the radio message in question was transmitted on November 29 of the same year. The fact that the same key was already in use 10 days earlier conflicts with the previous understanding. Astra did not treat this time gap as a problem and widened the effective range of the key for verification. It is speculated that past decryption attempts were constrained by the assumption of December 9 onward. The mistaken premise may have narrowed the search scope and caused years of failure. In cipher history, records of key change dates are fragmentary, making misinterpretations likely. There is room to see AI’s unbiased testing of candidates as the breakthrough. It is a discovery prompting a reexamination of operational history, with hopes for application to other unsolved messages.

Growing Use of AI in Historical Decryption

Decryption of historical ciphers by language models has been steadily increasing in recent years. Classical ciphers like ADFGVX are a field requiring both computational power and/or linguistic knowledge. Previously, progress relied on expert knowledge and manual trial and error, and was slow. Large language models handle older German spellings and abbreviations and evaluate candidates at high speed. Combined with image recognition, they can process everything from transcription of original materials to decryption in one workflow. If cross-checking against military records can be automated as in this case, the verification burden is greatly reduced. Unread cipher documents held by archives remain in various parts of the world. The use of AI is expected to raise productivity in historical research and expand the use of buried primary sources. Meanwhile, to prevent the introduction of misreadings, human source criticism remains essential. Collaboration between technology and history is assessed as the foundation for future decryption research.

Editorial Opinion

On short-term impact. This case is seen as encouraging the application of language models to classical ciphers. Reverification by archives and enthusiasts is expected to increase over the next three to six months. Challenges to the remaining items on the unsolved cipher list will intensify, and a shared platform for decryption reports is assessed to be developed.

On the long-term perspective. End-to-end processing from transcription of historical materials to decryption and record matching is seen as becoming established. Collaboration between historical research and AI technology will advance, raising the value of publicly released archival materials. Use in education and exhibitions is likely to expand within one to three years.

Questions from the editorial team. Why was the established theory on the key’s start date off by 10 days. Whether due to missing records or interpretive error, verification is considered necessary. While the AI output was corroborated by historical facts, room for misreading is seen as remaining. Reproduction experiments by third parties and source criticism are called for.

References

• “ ChatGPT-6 Astra cracks 108-year-old unsolved WWI German code for the first time — radio message sharing enemy movement intelligence had evaded decoding, 1918 Crimean fleet warning verified against HMS Canterbury logs “, by Mark Tyson — Tom’s Hardware, 2026-09-19T15:02:02.000Z (ARR)
• Source URL: https://www.tomshardware.com/tech-industry/artificial-intelligence/chatgpt-6-astra-cracks-108-year-old-unsolved-wwi-german-code-for-the-first-time-radio-message-sharing-enemy-movement-intelligence-had-evaded-decoding-1918-crimean-fleet-warning-verified-against-hms-canterbury-logs