iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Researcher Carter Church says GPT-6 Astra helped him produce a working reading of a cipher letter to French general Auguste de Marmont in about six hours of model execution. His account describes a detailed, checkable workflow, but the reading is not independently confirmed by OpenAI, and Church identifies five symbols that remain uncertain. The evidence supports a reported case study—not proof that the model alone solved the cipher or that AI can generally crack historical codes.
What was the Marmont cipher?
The document is a letter addressed to General Marmont, dated by Church to late March 1809. The reproduced letter begins with plain French and continues in 24 rows of cipher. Church says the image available to him was a 1,202 × 1,836-pixel plate reproduced in a 1969 issue of Revue historique des Armées, rather than a newly digitized original: Church’s account and solution package.
Church reports that the cipher contains 1,300 units and 155 distinct signs. The letter had a partial key: cryptology historian Daniel Tant’s table supplied 33 known letter values, which Church says accounted for 435 units. That gave the solving process some footholds, but left most of the cipher to be worked out.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Fix the driver behind crashes, sound loss and screen glitches3Repair Windows errors before they cause bigger problemsHow did Church use GPT-6 Astra?
Church describes a sequence combining human preparation, known cipher information, and computational search. The model’s reported role was part of this process; the account does not establish that it independently deciphered the letter from an untouched image.
#1 Best Overall
- Transcribe the reproduction. Church converted the journal plate into a working transcription of the cipher signs.
- Group likely glyph variants. Because handwritten signs can vary in appearance, he grouped forms he believed represented the same sign.
- Apply known values and a known word. He used the 33 letter values in Tant’s partial table and the plaintext word “CONSEQUENT” as constraints.
- Search for remaining assignments. Church says the solver used simulated annealing, a search method, together with French three-, four-, and five-character sequence statistics drawn from works by Victor Hugo, Alexandre Dumas, and Marmont’s memoirs.
- Check candidates against the image. Proposed assignments were compared with sign occurrences in the reproduced plate rather than accepted solely because they made plausible French.
Church reports about six hours of GPT-6 Astra execution time. That figure describes model execution in his account; it is not a measured total for all human preparation, research, transcription, or verification, and the cited sources do not provide an independently audited timing.
What does the decoded letter say?
The proposed plaintext is a military briefing describing French and allied positions and forces before Austria’s 1809 campaign. Its significance is that the reported reading corresponds to information in a separate historical document. Church compares it with Napoleon’s letter to Eugène dated 16 March 1809, which directs Eugène to send Marmont the same force dispositions in cipher. In Church’s account, the sequence of details, locations, and troop figures provides a cross-check on the proposed reading.
Rank #2
The historical match is evidence in support of the solution, but it should not be confused with independent scholarly validation of every deciphered symbol. Church says 29 signs represented whole words and lists five unresolved signs for which he gives best guesses. Accordingly, “working reading” is more precise than “definitive, error-free translation.”
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallCrashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteHow strong is the evidence for the claim?
The primary account is Church’s own write-up, published 18 September 2026. Coverage by 36Kr on 2 October 2026 and a Reb Babel roundup on 3 October repeat the claim through secondary reporting; they do not amount to separate technical replications. The available sources do not show OpenAI independently confirming this specific result.
Church makes the result more inspectable by providing a key, transcription, row-by-row decoding, and scripts intended to verify and regenerate the reading offline. He also reports that a rerun without Napoleonic texts and Marmont’s memoirs in the language model recovered the same reading. These details help readers examine the method, but they remain claims in Church’s account, not a controlled external evaluation.
- Reported: GPT-6 Astra execution took about six hours, according to Church.
- Not established: that the model worked alone, that every sign has an unambiguous reading, or that the result has been independently validated by OpenAI or outside researchers.
- Not demonstrated by one case: a general ability for AI models to solve historical ciphers, or superiority to human cryptanalysts or other models.
For readers who want to inspect the primary material, Church’s account includes the working materials: Breaking the Marmont Cipher 1809. The secondary retelling is available from 36Kr; the later summary is at Reb Babel.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
Free tools Windows power users keep installed
One-click scans. No signup required.
