Sunday 4 October 2026
Afternoon EditionRSS
ZOTPAPER

News without the noise

ZOTPAPER
AI

OpenAI's GPT-6 Astra Cracks 108-Year-Old Unsolved WWI German Cipher

A developer used the AI model to break an ADFGVX-encoded radio message from 1918, revealing a warning about Allied naval movements near Sevastopol.

By LineZotpaper
Published
Read Time2 min
A 108-year-old encrypted German World War I radio message, long considered unsolved, has been deciphered for the first time using OpenAI's GPT-6 Astra AI model, according to a developer named Prinz. The decoded message, which had stumped cryptographers for over a century, warns German high command of an English cruiser arriving at Sevastopol and an Allied squadron following on the 26th.

The message was encoded using the ADFGVX method, a complex German cipher from World War I that relied on a grid of letters shifted by a keyword. Prinz, the developer who claims the breakthrough, had taken the encrypted text from a list of 50 unsolved ciphers maintained by the German science blog Scienceblogs.de. The message was addressed to the German High Command and likely intended for an admiral or naval command.

Using GPT-6 Astra, Prinz cracked the code with the keyword "TRUPPENVERSCHIEBUNG" (German for "troop movement"). The decoded text reads: "EIN ENGLISCHER KREUZER EINLIEG X SEWASTOPOL X S4STEN X EIN GESCHWADER DER X ALLIIERTEN FOLGT 26STEN X," which translates to: "AN ENGLISH CRUISER ARRIVED AT SEVASTOPOL ON THE ?4TH AN ALLIED SQUADRON FOLLOWS ON THE 26TH."

The cipher's solution was verified against historical records, specifically HMS Canterbury logs, according to Prinz. The message had evaded decoding since its transmission in 1918, and its content aligns with known Allied naval operations in the Crimean region during the final year of the war. The developer shared the result on X (formerly Twitter), posting a screenshot of the decoded message from GPT-6 Astra.

The achievement highlights the growing role of large language models in historical analysis and cryptanalysis, though some experts caution that AI-driven decryption may rely on pattern recognition rather than traditional cryptographic methods.

§

Analysis

Why This Matters

  • Demonstrates the expanding utility of advanced AI in specialized fields like cryptanalysis, where unsolved historical ciphers may now yield to large language models.
  • The breakthrough could inspire renewed efforts to decode other long-standing historical ciphers, potentially unlocking new wartime records.
  • Raises questions about the reliability of AI-decoded messages and the need for traditional verification methods, given the historical significance.

Background

The ADFGVX cipher was a German military field cipher introduced in 1918 during World War I, designed for secure radio communication. It used a 6x6 grid encoding letters and digits, with a second transposition step based on a keyword, making it extremely difficult to break without the key. Many such messages from the war remain unsolved, often due to lost keys. Scienceblogs.de maintains a prominent list of 50 unsolved WWI ciphers, from which Prinz selected this particular message.

Key Perspectives

[Developer Prinz]: Claims credit for successfully decoding a century-old cipher using GPT-6 Astra, viewing it as a practical demonstration of the AI's advanced reasoning capabilities for complex puzzles. [Historians and cryptographers]: Likely to welcome the potential for unlocking lost intelligence but will demand rigorous verification and may question whether AI-driven pattern recognition constitutes true cryptanalysis. [Skeptics]: May argue that without access to the original key or multiple independent confirmations, the decoding could be a coincidental match rather than a legitimate solution, especially given the messiness of historical records.

What to Watch

  • Whether other unsolved ciphers from the list are subsequently decoded using similar AI methods.
  • Historical scrutiny of the HMS Canterbury logs to confirm the decoded content's accuracy.
  • OpenAI's response to this use case and any additional context it may provide about GPT-6 Astra's capabilities in long-standing analytical tasks.

Sources

Zotpaper

Written by software from the reporting listed above, scored by an automated standards desk, and published without a person reading it first. If something here is wrong, tell the editor and it will be put right.

How we workReport an errorSubscribe