For more than a century, one World War I German radio message sat inside a collection of unsolved ADFGVX cryptograms. The 170-symbol transmission was dated November 27, 1918, and despite decades of interest in the surviving material, its meaning had remained unclear.

Now, a reported experiment involving OpenAI’s GPT-6 Astra has produced a German plaintext that appears to line up with surviving Royal Navy records. The result is intriguing not simply because an old cipher was deciphered, but because the proposed reading points toward real naval movements around Sevastopol — while still leaving important questions about exactly how the solution was reached.

More AI Capability Watch: GPT-6 Astra’s ability to work through complex information keeps turning up in unexpected experiments. For more, check out our story on GPT-6 Astra sparking a “simulation within a simulation” experiment and whether AI just made Nick Bostrom’s theory feel real.

How GPT-6 Astra decoded the 1918 German ADFGVX radio message

A cipher that survived more than 100 years

The ciphertext belongs to a collection of German ADFGVX radio messages from World War I. ScienceBlogs’ Klausis Krypto Kolumne previously featured the material in its “Top 50 unsolved encrypted messages” series.

Its entry on the World War I ADFGVX material said approximately 20 messages from 1918 remained unsolved at the time of publication. The wider collection came from material preserved by U.S. Army officer James Rives Childs.

Many other German ADFGVX messages from the same period have since been deciphered. Cryptanalyst George Lasry and collaborators developed computerized techniques that solved large portions of the surviving material. The ScienceBlogs archive notes that hundreds of ADFGVX cryptograms from July through December 1918 were preserved and that many were later solved using modern cryptanalysis.

The particular message examined by the developer publishing under the name Prinz was among those that had continued to resist a solution.

More Ancient and Historical Mysteries: Long-buried secrets from history keep resurfacing through modern research methods. For more, read our feature on the ancient “poop explosion” that may have helped fuel the Cambrian Explosion.

What made the ADFGVX system difficult?

ADFGVX was a German military cipher used during the final period of World War I. Its name comes from six letters: A, D, F, G, V and X.

Those six symbols functioned as coordinates within a six-by-six grid. Letters and numbers could be represented by assigning each character a pair of those symbols.

But that was only the first stage.

The resulting ADFGVX symbols were also subjected to a columnar transposition. The symbols were rearranged according to a keyword, which meant that identifying the correct key was central to reversing the encryption.

The system was used for German radio communications between Germany and outposts and stations in regions including the Balkans, the Black Sea and the Middle East.

That combination of substitution and transposition made ADFGVX considerably more difficult to solve than a straightforward substitution cipher.

More Brain and Cognition Frontiers: Connecting scattered pieces of information into something coherent isn’t limited to code-breaking. For more, check out our story on Ray Kurzweil joining a startup trying to hack the brain without surgery.

The 19-letter keyword that changed the investigation

The reported breakthrough centred on one historical keyword:

“TRUPPENVERSCHIEBUNG”

The German word roughly means “troop movement” or “troop redeployment.”

According to Prinz’s account, the word appears in J. Rives Childs’ historical material on German military ciphers. The relevant material is on pages 214 and 215.

Astra used the keyword to reconstruct the transposition arrangement and work backward through the 170-symbol ciphertext.

The keyword contains 19 letters. The ciphertext could therefore be arranged around a 19-column structure, allowing the columnar transposition to be reversed. Once the columns were restored, the ADFGVX coordinate system could be applied to recover German plaintext.

This detail changes how the reported achievement should be understood.

The model did not independently discover an unknown cipher system from a completely untouched ciphertext. The historical key already existed in surviving reference material. The reported contribution was Astra’s ability to connect that documented keyword to this particular unsolved message and then execute the relevant decryption process.

More Careful Verification Stories: Resisting the temptation to round uncertainty into false confidence is a hallmark of solid research. For more, read our feature on the North Carolina petting zoo rabies scare after baby goats tested positive following Durham events.

The German plaintext revealed a naval warning

According to Prinz’s published account, Astra recovered this German plaintext:

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

The recovered text contains an uncertain character around the date. That means the number cannot be read with complete certainty directly from the decrypted material.

The reported English translation reads:

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

That question mark is significant.

The proposed solution does not cleanly establish the first digit of the date. Instead of silently converting the uncertain character into a definite number, the experiment compared the message against independent historical information.

That comparison produced the most striking part of the story.

More Records That Confirm the Unexpected: Independent documentation lining up with a surprising claim always strengthens the case. For more, check out our story on Mike Myers’ $25K Cat in the Hat costume being called “haunted,” and whether there’s really something inside it.

HMS Canterbury provides an independent historical cross-check

The British light cruiser HMS Canterbury was operating in the region in November 1918. Its surviving ship’s log records that the cruiser left Constantinople for Sevastopol on November 23 and arrived at Sevastopol on November 24, 1918.

The same record states that an Allied Squadron arrived on November 26, 1918.

Those dates closely match the sequence described by the recovered German message.

Date	Recorded event
November 23, 1918	HMS Canterbury left Constantinople for Sevastopol.
November 24, 1918	HMS Canterbury arrived at Sevastopol.
November 25, 1918	British personnel landed to take control of German vessels.
November 26, 1918	The Allied Squadron arrived at Sevastopol.
November 27, 1918	The German radio message examined in the reported experiment was transmitted.

The match provides a concrete external consistency check for the proposed plaintext.

It also offers a plausible explanation for the uncertain character in the recovered date. If the message was referring to HMS Canterbury, the intended date appears consistent with the 24th.

However, the original uncertainty still matters. The ciphertext’s recovered character should remain uncertain rather than being presented as a definite “24” solely because the historical record fits.

More Unresolved Puzzles Worth Watching: Mysteries with strange, unexplained wrinkles tend to stick around in the public imagination. For more, read our story on the Brentwood “serial pooper” mystery taking a darker turn after an alleged death threat.

A surprising problem remains with the historical keyword

There is another complication.

The keyword “TRUPPENVERSCHIEBUNG” is documented in historical material as being associated with a period beginning December 9, 1918, according to the account of the experiment.

The ciphertext in question is dated November 27, 1918.

That creates a chronological discrepancy of several weeks.

It also offers a possible reason why the message had remained unresolved. Researchers who assumed that the documented key could not have been used before December 9 would have had little reason to test it against a November 27 transmission.

Still, the timing issue cannot simply be ignored.

A later documented start date does not by itself prove that the keyword could not have been used earlier. Historical records can be incomplete. Operational practices can also differ from surviving documentation.

At the same time, the discrepancy means the proposed solution warrants careful independent examination.

More AI-Assisted Breakthroughs Under Scrutiny: Impressive AI results still need independent verification before they’re treated as settled fact. For more, check out our story on Sam Altman’s claim that 38,000 ChatGPT queries equal one almond’s water use, and whether anyone can actually prove it.

What Astra actually demonstrated

The distinction between finding a key and using a known historical key is central to the story.

The keyword was already present in surviving historical material. Human researchers had preserved that information for decades.

Astra’s reported achievement was recognizing that “TRUPPENVERSCHIEBUNG” could solve this particular unresolved ciphertext, reconstructing the transposition, applying the ADFGVX system and producing German plaintext that made historical sense.

That is an interesting example of AI-assisted research and reasoning because it connects several different forms of information.

The model reportedly brought together the surviving ciphertext, knowledge of ADFGVX, a published historical keyword, columnar-transposition mechanics, German-language plaintext and independent naval records.

But the available reporting does not establish that the result has undergone formal peer-reviewed cryptanalysis by independent specialists.

That makes the most accurate description a reported AI-assisted decryption, rather than an unquestionable independent cryptanalytic breakthrough.

More Wartime History Rediscovered: Records from major conflicts continue to surface new details decades after the fact. For more, read our feature on the Renoir museum heist, where two paintings remain missing after thieves cut through frames in a French Riviera raid.

Why the historical match is so important

A plausible-looking plaintext does not automatically prove that a cryptographic solution is correct.

A wrong key can sometimes generate fragments that appear meaningful. That risk becomes particularly important when researchers test possible arrangements against historical information.

Here, the proposed plaintext contains specific information about an English cruiser reaching Sevastopol and an Allied squadron following on the 26th.

The surviving HMS Canterbury log independently records the cruiser reaching Sevastopol on November 24 and the Allied Squadron arriving on November 26.

That does not independently prove every step of Astra’s reasoning. It does, however, provide a concrete historical consistency check for the proposed reading.

The correspondence is especially notable because the uncertain date in the decrypted message can plausibly be reconciled with the documented November 24 arrival.

More Chaotic Historical Turning Points: The final days of major world events often produced the most consequential, hard-to-trace decisions. For more, check out our story on Denmark’s biggest Viking silver hoard, hidden under a backyard until a terrace project exposed it.

A glimpse into late-war naval intelligence

The decoded communication appears to concern Allied naval movements around Sevastopol during the chaotic closing period of World War I.

The transmission came only days after the Armistice of November 11, 1918. At that point, the military and political situation around the Black Sea region was changing rapidly.

Information about an English cruiser and an approaching Allied squadron would have been relevant to German authorities monitoring movements in the area.

The exact recipient remains less certain.

The available account describes the message as a German communication directed toward the German High Command and potentially relevant to naval command. However, the surviving material does not justify treating the identity of a particular admiral as established fact.

That distinction is important because the decrypted content may provide useful historical intelligence without answering every question about how the message was distributed or who specifically received it.

More Capability Milestones and Their Limits: New AI models keep raising the bar, while researchers keep reminding everyone to verify before celebrating. For more, read our story on ex-Google DeepMind researcher Bilal Chughtai quitting over AI safety fears and his warning that “AI has the potential to kill us all”.

GPT-6 Astra’s recent launch adds another layer

The experiment comes only weeks after OpenAI released GPT-6 Astra.

OpenAI announced Astra on September 3, 2026, describing it as its most capable broadly deployed model at launch. The company says the model was designed for advanced computer use, browsing, software engineering, cybersecurity, science and professional work.

OpenAI’s safety documentation also describes Astra as a significant increase in cybersecurity capability and says the model reached the company’s “Critical” threshold under its Preparedness Framework.

The historical cipher demonstration is therefore one of several recent examples being used to illustrate Astra’s ability to work across complex information and reasoning tasks.

But the cipher case also shows why those capabilities still need careful checking.

A model can produce a coherent answer. Researchers still need to establish that the solution genuinely follows from the original evidence and is not the result of an incorrect assumption, an overlooked alternative key or information accidentally supplied through the model’s source material.

More Curious Corners of the Internet: Not every fascinating story requires wartime intrigue — some are just delightfully odd. For more, check out our roundup of the 2026 Ig Nobel Prizes, featuring cockroach milk, splash-free urinals and 1,000 pairs of underwear.

A 108-year-old cipher mystery now has a compelling proposed solution

The reported GPT-6 Astra experiment offers a coherent reading of a World War I German ADFGVX message that had remained unresolved for decades.

The proposed plaintext describes an English cruiser reaching Sevastopol and an Allied squadron following on the 26th. The surviving HMS Canterbury records independently document the cruiser arriving at Sevastopol on November 24, 1918, followed by the Allied Squadron on November 26.

The identification of “TRUPPENVERSCHIEBUNG” also gives researchers a concrete route for reproducing and testing the proposed solution.

Yet important questions remain. The documented timing of the keyword begins on December 9, 1918, while the ciphertext is dated November 27. The recovered date contains an uncertain character. The available reporting also does not establish formal independent peer-reviewed cryptanalytic confirmation.

For now, the most defensible description is that GPT-6 Astra has reportedly provided a compelling AI-assisted solution to a previously unresolved 1918 German ADFGVX message. Its proposed reading is strengthened by surviving Royal Navy records, while the outstanding historical and cryptanalytic questions leave room for further independent examination.

Source and research disclaimer

This article has been prepared from thorough research of the sources provided with the original material, including OpenAI’s official information about GPT-6 Astra, the ScienceBlogs/Klausis Krypto Kolumne archive, Prinz’s published account of the reported decryption, Naval-History.net’s HMS Canterbury service record and the Tom’s Hardware report. The article reflects the information available from those sources and preserves the distinctions between documented historical records, the reported AI-assisted solution and unresolved questions. It does not imply independent verification beyond what the provided material supports.

Sources

• OpenAI — GPT-6 Astra: A new generation of intelligence
• OpenAI — Safety overview: GPT-6 Astra
• ScienceBlogs / Klausis Krypto Kolumne — The Top 50 unsolved encrypted messages: Unsolved ADFGVX cryptograms from World War 1
• Prinz / reported GPT-6 Astra cipher analysis
• Naval-History.net — HMS Canterbury, light cruiser, World War I service log
• Tom’s Hardware — Report on GPT-6 Astra and the 1918 German cipher

Featured Image Credit:

• Sami Abdullah on Pexels