AIToday
Large Language ModelsTHE DECODERPublished: Sep 17, 2026, 19:01 JST

GPT-6 Astra cracks 83-year-old Enigma message, says Bloomberg developer

GPT-6 Astra cracks 83-year-old Enigma message, says Bloomberg developer

3 Key Points

  1. What happened

    Carter Leffen, a product development coach at Bloomberg LP, says OpenAI's GPT-6 Astra worked about ten hours to decrypt an 82-character Wehrmacht Enigma message from 1941 that had been listed as unsolved for 83 years.

  2. Why it matters

    A single AI model handling archive searches, an Enigma simulator, cryptanalysis code and key tests suggests AI agents can carry a multi-step research process that previously demanded scarce specialist skills.

  3. What to watch

    Whether the decryption holds up to independent review will depend on experts evaluating the published materials, since the released packages verify the calculations but not the message's historical identity. Leffen says he put "99 times more effort into building the website" than into cracking the code.

WHO IT HITSHistorians, cryptanalysts, and archivists sifting intercepted wartime material may gain a faster first-pass tool for stalled decryptions, while researchers using AI agents for multi-step analysis get a public test case — assuming independent experts validate the result.

Not sure about something? Ask the AI

Summaries like this, in your inbox every morning.

Context & Analysis

The message itself is modest — a German soldier in 1941 reporting he was in Rosenow and asking for his march route. What makes the case notable is the method. Leffen and several specialized AI agents ran the search end to end: combing historical archives, comparing uncertain letters, building an Enigma simulator, writing cryptanalysis code, and cross-checking results. The Enigma offered roughly 159 quintillion possible daily settings, so brute force was never viable; the winning shortcut was a town name, "Rosenow," that also appeared in an already-decrypted message from the same day.

The work builds on prior cryptanalytic contributions by Frode Weierud, Geoff Sullivan, and Olaf Ostwald. Leffen points to the typos in the recovered text — "BTTE" instead of "BITTE" — as evidence of authenticity, since a fabricated answer would more likely be error-free.

Whether this holds up hinges on independent review of the published materials. As the article notes, the released packages verify the calculations but not the message's historical identity or that the solution is the only possible one — so experts will decide whether this becomes a template for AI-assisted cryptanalysis or a well-documented one-off.

FAQ
How did GPT-6 Astra break the Enigma message?
The breakthrough came from a known town name: an already-decrypted message from the same day contained "Rosenow" twice, and Leffen guessed it would also appear in the unsolved message. A well-known Enigma weakness — no letter ever encrypts as itself — let the team rule out many positions.
Can I verify the decryption myself?
All code, search data, and a working 3D Enigma simulator are available for download on the project page. The packages verify the calculations and replicate the key search sequence, but they don't prove the message's historical identity or that the solution is the only possible one.
How much work did Leffen say the codebreaking took?
By his own account, the codebreaking wasn't even the hard part. He says he put "99 times more effort into building the website that describes the problem and the solution than into actually cracking the code."

Get the latest Large Language Models news every morning

For example, today's edition would include:

  • Discover Labs: AI search makes the LLM the new website visitorPractical AI · 2h ago
  • OpenAI says internal model cracked Navier–StokesLast Week in AI · 2h ago
  • Huawei unveils OceanStor M900 at Huawei Connect 2026DIGITIMES Asia · 5h ago

AI-summarized, only the topics you pick — one digest a day via Email, Slack, or Discord.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. Q&As are published on this page for other readers too.

Related Articles

Next articleMary Barra: trades beat white-collar jobs on AI-proofing