GPT-6 Astra, a large language model featured in the source, successfully decoded a World War I German radio message dated November 27, 1918, that had remained unsolved despite hundreds of similar messages being broken using known German keys. This cipher is one of fifty unsolved ciphers listed on the Scienceblogs.de German science blogging portal, a collection described in the source as ranging from serial‑killer cryptograms to the Voynich manuscript. The solution employed the ADFGVX field cipher, a German Army encryption method from WWI that combines a modified Polybius square with columnar transposition.

Using the encryption key 'TRUPPENVERSCHIEBUNG', which the source identifies as being documented on pages 214‑215 of J. Rives Childs's reference work 'The History and Principles of German Military Ciphers, 1914–1918', the model reconstructed the ADFGVX table, rearranged the key alphabetically, and applied columnar transposition to recover the German plaintext: 'EIN ENGLISCHER KREUZER EINLIEG X SEWASTOPOL X S4STEN X EIN GESCHWADER DER X ALLIIERTEN FOLGT 26STEN X'. The source translates this as 'AN ENGLISH CRUISER ARRIVED AT SEVASTOPOL ON THE ?4TH AN ALLIED SQUADRON FOLLOWS ON THE 26TH', noting the uncertain character likely represents a numeral. To corroborate the historical claim, the source states that the model consulted the original log entries of the English cruiser HMS Canterbury, which recorded the ship's arrival in Sevastopol on November 24, 1918, and noted that an Allied squadron followed on November 26, matching the message’s prediction.

Nevertheless, the source highlights a chronological tension: the radio message was transmitted on November 27, 1918, whereas the cited historical reference indicates that the key 'TRUPPENVERSCHIEBUNG' was not introduced by German forces until December 9, 1918, over a week later. This mismatch raises unresolved questions—whether the key was actually deployed earlier than recorded, whether the message date is inaccurate, or whether another cryptographic explanation accounts for why this particular cipher resisted solution for over a century. The author concludes that, to their knowledge, this specific Nov 27, 1918 message had not been decoded before, presenting the model’s result as a novel demonstration of large‑language‑model capability in historical cryptanalysis when guided by appropriate reference material.