An encrypted Nazi military message sat unread for decades.

GPT-6 Astra helped break it in roughly 10 hours.

The 82-letter German Army transmission was sent on July 10, 1941, while Adolf Hitler’s forces were driving deep into the Soviet Union.

Its contents were not a grand strategy or a secret order from Hitler’s headquarters. They were more immediate: a field message asking for a route of march, reporting a position in Rosenow and demanding an immediate reply by radio.

That modest plaintext had remained hidden behind one of the most notorious encryption systems of World War II.

ADVERTISEMENT

Carter Leffen, the researcher who directed the project, described what Astra did during the recovery:

The achievement spread quickly after Polymarket highlighted it Thursday:

The most important detail is that this was not a chatbot making a clever guess and declaring victory.

The MVUEH recovery exhibit and validation record lays out the work in unusually specific terms.

Leffen led the investigation and made the key decisions. Astra and specialist agents searched historical material, compared uncertain letters in two surviving copies, built an Enigma simulator, wrote cryptanalysis code, ran controlled experiments and tested competing settings.

The project did not begin with a blind brute-force attack. The standard machine model left an estimated 159 quintillion daily configurations before other constraints were applied, and early approaches failed.

The team worked across September 14 and 15, using a faint outgoing facsimile, a separately published received transcription and a recorded message header. The two surviving copies disagreed at several positions, so the investigation recorded possible letters before searching instead of changing inconvenient characters after a promising key appeared.

The final reconstruction changed three characters from an earlier native transcription and identified eight differences from the received copy. Every accepted letter came from an alternative that had already been documented, giving reviewers a concrete trail to challenge.

The breakthrough came from a related message.

ADVERTISEMENT

A previously solved transmission identified as SIPVX included the place name ROSENOW twice in succession. The team treated that 14-letter sequence as a possible “crib” — a suspected piece of plaintext that could be tested against different positions in the encrypted message.

That was a major hint, but it was not the answer.

The crib accounted for only 14 letters. The other 68 letters still had to produce coherent German, and a separately recorded header had to agree with the recovered machine settings.

One candidate survived those tests.

The recovered key used Enigma I with reflector B, rotor order II-V-III, ring settings H-M-F and a body start of R-W-D. The plaintext around the repeated place name formed a connected request for marching instructions and an immediate radio response.

Grok posted the rough English meaning alongside part of the literal German text:

The errors in the recovered German matter.

The text contains “BTTE” where a writer would normally use “BITTE,” along with other apparent operator mistakes. The project retained those irregularities instead of silently polishing them into a cleaner message.

That is what serious verification looks like: preserve the awkward evidence, disclose the guessed portion and force the proposed key to explain everything around it.

ADVERTISEMENT

The investigation also examined alternate readings of faint characters in the surviving documents. Three positions changed from an earlier transcription, and the team says every selected alternative had been documented before the key was discovered.

After the initial recovery, a separate search used a phrase learned from the solution and checked all 1,080 batches. It produced thousands of physical key candidates, but only one matched the original recorded header.

The full investigation covered 43,016 batches and preserved rival candidates for review.

That level of disclosure is especially important when AI is involved. A fluent model can make a wrong answer sound inevitable.

Cryptography does not care how confident the prose sounds. The settings either reproduce the ciphertext and header or they do not.

The external corpus now reflects the result.

Frode Weierud’s CryptoCellar German Army Enigma list identifies the transmission as MVUEH, dated July 10, 1941, and records it as broken by Carter Leffen on September 14, 2026.

The archive places it inside a much larger collection of German Army traffic. Its July 1941 index contains dozens of ciphertexts, dates, message identifiers and body lengths, allowing researchers to compare MVUEH with traffic sent around the same time instead of treating it as an isolated puzzle.

MVUEH appears as message 172 and is listed with an 82-letter encrypted body. The corpus also preserves its archival reference and links to the surviving material used in the recovery.

ADVERTISEMENT

That independent record matters because it predates the social-media attention. It shows where the ciphertext came from, how specialists tracked its status and when the new result entered the public corpus.

There are still honest limits.

The recovery team does not claim that every faint handwritten character is visually certain. The exact Rosenow referenced in the message has not been established, the final signature is tentative and the relationship between MVUEH and the related solved message remains unproved.

The team also declined to make a sweeping world-first claim because older corpus labels can be ambiguous.

None of that erases the core result: a documented key reproduces a coherent 1941 German military message, matches its surviving header and has been added to the specialist corpus as solved.

It is also worth remembering whose shoulders this work stands on.

The National Security Agency’s history of Marian Rejewski credits the Polish mathematician with the initial analysis that made exploitation of German Enigma possible. His higher-algebraic attack solved a problem that had defeated French and British cryptanalysts for years.

Rejewski began attacking the machine in 1932. Working with Jerzy Różycki and Henryk Zygalski, he developed methods for recovering Enigma’s daily settings and helped create the bomba, a machine the British later developed further.

Those breakthroughs, passed to France and Britain before Germany invaded Poland, became part of the foundation for the Allied cryptanalytic campaign.

The NSA credits that Polish work with giving Britain and the United States the opening they needed to exploit German military communications during the war.

Astra’s result belongs to a different era and a different kind of problem. The Allies had to break live military traffic fast enough to affect a war.

This project returned to one short archival message that had resisted modern hobbyists and researchers.

But the new work still points to something consequential.

The model searched archives, worked through bad scans and uncertain transcriptions, wrote code, ran simulations, absorbed failed experiments and helped assemble independent checks into a reproducible cryptanalytic case.

For 85 years, MVUEH was a block of letters.

Now it is a voice from Hitler’s war, asking where to march next.

This is a Guest Post from our friends over at WLTReport. View the original article here.

 

Join The Conversation. Leave a Comment.