An artificial intelligence model has successfully decrypted a 107-year-old World War I German military radio message that had remained unreadable since November 1918. The breakthrough relies on OpenAI’s GPT-6 Astra and historical logs from a British cruiser, according to recent blogger reports.
A century-old cryptographic puzzle from the final days of the First World War has finally yielded its secrets. A mysterious German military transmission sent on November 27, 1918—just days after the armistice ended the global conflict—remained an impenetrable string of cipher text for more than a hundred years. Now, an OpenAI artificial intelligence model has broken the code, translating a message that historically resisted deciphering.
The decryption effort was published online by a writer using the pseudonym prinz
on an internet forum, who identified the software responsible as the GPT-6 Astra model. While hundreds of similar wartime dispatches had previously been cracked by professional cryptologists such as George Lasry, more than a dozen messages from the era defied all attempts. The newly resolved ciphertext sat on a list of 50 unsolved codes maintained by the German science portal scienceblogs.de.
The ADFGVX Cipher and the Hidden Keyword
The historic dispatch was secured using the notorious ADFGVX cipher, a complex field-encryption method deployed by the German military during World War I. The system relied on a six-by-six grid where the axes were labeled with the letters A, D, F, G, V, and X, yielding 36 distinct cells to accommodate the alphabet and numerals. Senders scrambled the letters by inserting a secret keyword that was periodically changed.
Without knowing the exact keyword, reconstructing the pairs proved nearly impossible for modern researchers examining the remaining archives. In this instance, the AI model employed the keyword Truppenverschiebung
—a German term meaning troop movement—to unravel the matrix.
Historical records present a chronological puzzle regarding this specific key. However, historical documentation previously indicated that this particular keyword only came into regular military use beginning on December 9, 1918. Because the encrypted radio message was broadcast nearly two weeks prior to that date, earlier human analysts likely abandoned their attempts, assuming the keyword would not match the timeline.
Decoded Content and Royal Navy Logbooks
Once the AI applied the matrix, the resulting plaintext emerged in structured segments separated by placeholders. The translation reads according to the reporting that an English cruiser had entered Sevastopol on the 24th, followed by a squadron of the allies on the 26th. Translated into plain English, the message reports that an English cruiser had entered the port of Sevastopol in Crimea, and that an allied squadron would follow on the 26th.
An intriguing anomaly appears in the date of the cruiser’s arrival: the first digit is replaced by the letter S, leaving only the number 4 visible, which researchers interpret as a typo for the 24th. To verify the accuracy of the decrypted text, the system cross-referenced the details against historical maritime archives, specifically the logbook of the British cruiser HMS Canterbury.
Naval records confirm that the HMS Canterbury arrived in Sevastopol on November 24, 1918. Exactly two days later, the ship’s logbook records the subsequent arrival of an allied squadron, aligning precisely with the timeline transmitted in the intercepted German wireless dispatch.
Methodology and Verification Limits
The discovery highlights how modern machine learning models can process historical cryptographic challenges by rapidly testing candidate keys against vast datasets of contemporaneous records. By pairing brute-force matrix permutation with historical cross-checking, the software bridged the gap between an unreadable archive and verified naval movements.
Despite the precise match with the British naval records, independent verification remains incomplete. The decryption has been publicized through independent blogging channels rather than through formal peer-reviewed academic publication. Professional cryptographers and historians have yet to issue a formal counter-analysis or verification of the methodology, leaving the finding classified as a compelling amateur breakthrough that demonstrates the evolving capabilities of large-scale language and reasoning models.
