GPT-6 Astra Solves Unsolved WWI German Radio Cipher

GPT-6 Astra Decodes WWI ADFGVX Cipher

GPT-6 Astra has successfully decoded a World War I German radio message that had previously eluded cryptanalysis. The model identified the correct encryption key and verified the resulting plaintext against historical naval logs, solving a cipher transmitted on November 27, 1918.

The Decryption Process

The message was encoded using the ADFGVX cipher, a field cipher used by the German Army during WWI. This method employs a 6x6 grid (fractionation) and a columnar transposition based on a keyword.

To solve the cipher, GPT-6 Astra utilized the encryption word "TRUPPENVERSCHIEBUNG" (German for "troop shift"). The decryption process involved two primary steps:

Grid Mapping: The model used the keyword to generate a 6x6 substitution table where each letter is represented by a pair of ADFGVX coordinates.
Columnar Transposition: The keyword "TRUPPENVERSCHIEBUNG" was used to determine the order of columns for rearranging the encrypted symbols. The model calculated the column positions based on the alphabetical order of the letters in the keyword to retrieve the original sequence of coordinate pairs.

Resulting Plaintext and Historical Verification

The decoded German text reads: EIN ENGLISCHER KREUZER EINLIEG X SEWASTOPOL X S4STEN X EIN GESCHWADER DER X ALLIIERTEN FOLGT 26STEN X

Translated to English, the message states:

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

To validate the result, the model cross-referenced the plaintext with historical records. It found that the HMS Canterbury, an English cruiser, arrived in Sevastopol on November 24, 1918, and an allied squadron followed on November 26, as documented in the ship's original logs.

Why the Cipher Remained Unsolved

Astra hypothesized that the message remained unsolved because the key "TRUPPENVERSCHIEBUNG" was documented as being used starting on December 9, 1918. Because the message in question was transmitted on November 27, 1918, human analysts likely overlooked this key because it fell outside the expected timeframe for that specific encryption word.

Community Analysis and Critique

Following the announcement, technical discussions on Hacker News raised several points regarding the nature of the discovery and the capabilities of the model:

Methodological Skepticism

Some users questioned whether the model performed original cryptanalysis or simply retrieved a solution from its training data.

"Or, it found a human that solved it in the dataset and stole the solution."

Others suggested that since the list of German keys was seized after the war and is publicly available, the task was a matter of searching through known keys rather than breaking a new encryption method.

"TL;DR: all keys are known because the list was seized after the war... ChatGPT didn't need to brute-force the entire key, just pick the correct one from the list and identify the typo."

Complexity of the Task

Critics argued that the ADFGVX cipher is a relatively simple substitution and transposition cipher that could be solved by a dedicated human with a computer and a dictionary attack, provided they had the correct key list.

Verification of Sources

One user reported that when asked to describe the specific pages of the source cited by the model (J. Rives Childs's The History and Principles of German Military Ciphers), the model was unable to provide the content or find the book online, raising questions about the model's ability to accurately cite its sources in real-time.

Broader Implications

The event sparked a debate on the role of LLMs in cryptanalysis, with some viewing it as a sign of emerging AGI capabilities and others seeing it as an illustration of "fancy autocomplete" applying existing data to unsolved problems.