An open ledger · since September 2026

Unsolved historical ciphers

A shared ledger of historical ciphers: what has been deciphered, by whom and how, and why the rest still resist. The project uses frontier AI models, such as Anthropic’s Claude, to help decipher them, and takes contributions from anyone.

Daniel Bourdeau, Brice Dudout, George Lasry and contributors · posted · updated · contact

As seen in · · Klaus Schmeh’s blog

Philip II to Bernardino de Mendoza, 1589 · BnF Français 3641 f. 14 →

288 keys recovered286 from existing keys or decipherments1,387 ciphers still open

Results so far

Every entry by date, 1250–1991. Hover or tap a dot for the story; click a category under the bar to isolate it.

Entries by century, 1250–1991. Tap a century for its write-ups.

All 9 categories
28843% of attemptskeys recovered by cryptanalysis
breakdown122 from the ciphertext alone, 140 from plaintext beside the ciphertext, 26 from plaintext found elsewhere. 179 of them with outside contributors. In 10 of them the key was recovered but the text does not yet read.
28642% of attemptsobtained with a key or decipherment that already existed
breakdown130 with a known key, 48 with a key matched here from another document, 108 from a complete existing decipherment. In 15 of them the key was applied but the text does not yet read.
9915% of attemptsclosed without recovering the text; each write-up says why
96,395years of silence endedsummed over the 263 dated texts obtained in full (274 more partial), each counted from its date to 2026. The oldest is Ibn al-Qaysarani → the unnamed compiler of a Mamluk book of cipher verses (text c. 1330–1370; copy 15th c.).

706 entries, one per row of the results tables, each classed by how its text was obtained. 22 turned out not to be ciphers or could not be reached and 11 are in progress, which leaves 673 attempted; every percentage here is of those 673.Percentages are of the 673 attempted.

Counted from the results tables in the repository README by _build_stats.py; regenerated 16 November 2026. Rows are documents, not correspondents.

Three ways to help

Or check one of 230 readings waiting for an independent check, or send a cipher that is not listed. Every contribution is credited. How to contribute →

Twenty favourites: the funny, the historic, the ingenious, the hardest →

Latest findings

All 655 write-ups →

Browse the write-ups

All 655 write-ups, with filters and search → · the atlas · the key web · what the methods mean

The Unsolved Catalogue

1,387 undeciphered letters and series from 1200 to 1953. Each carries its shelfmark, the prior art checked, a difficulty class and the check that would confirm it is genuinely open; 333 have been looked at on the image.

Top of the queue

Open The Unsolved Catalogue →

Corrections and pointers to existing decipherments are welcome: a catalogue entry marked “unsolved” is often a literature search away from “solved in 1858”.

Contributors and sources