A four-century-old cryptogram embedded in a 1653 book by the Scottish writer Sir Thomas Urquhart has reportedly been decoded by Claude Fable 5.1, according to Vals AI, the evaluation company that ran the test. The puzzle, known as the Cyphral Distich, consists of 64 numbers arranged in two lines at the close of Urquhart’s “Logopandecteision.” It had resisted solvers since at least 1899, when it was posed as an open problem in the journal Notes and Queries, and it later appeared on cryptologist Klaus Schmeh’s list of the top 50 unsolved historical ciphers.
Vals AI’s account, published on its blog with no named byline, says earlier attempts tried homophonic and simple substitution plus frequency analysis, all of which failed to produce readable text. The company frames the miss as a matter of a missing clue rather than a failure of cryptanalytic skill.
According to the post, Fable 5.1 worked for 44 minutes and used 176,000 tokens without human intervention before landing on the method. The model reportedly noticed that the cipher sits directly after 32 numbered passages Urquhart called Proquiritations, and that an accompanying poem promises the reader will find “his own heart’s wishes” inside. Combining those two observations, Fable 5.1 is said to have derived the actual rule: take the number in position i of a line, go to Proquiritation number i, count forward that many words, and keep the first letter of the word reached. Applied across both lines, Vals AI reports the output reads “O GOD UPHOLD KING CHARLS THE SECOND AND MAKE HIM THE SUPREME RULER OF THIS LAND,” a prayer for the restoration of Charles II that lines up with Urquhart’s documented royalist politics.
The company points to internal consistency as evidence the answer is correct: each decoded line runs exactly 32 letters, and the lines rhyme on “and” and “land,” matching the “distich” form promised in the surrounding poem. Vals AI also says Fable 5.1 applied the same logic to a second, larger 285-number cryptogram in Urquhart’s 1652 work “The Jewel,” this time mapping numbers to page indices rather than paragraphs, and recovered all but nine letters of that text. The post lists several unresolved spots in that second solve, including one nine-letter run that does not resolve into readable English under any method the author says was tried.
None of this has been checked by an outside cryptography authority, and the write-up is Vals AI’s own account of its own test. The self-consistency checks (letter counts, rhyme, thematic fit with Urquhart’s biography) are persuasive but not the same as independent peer confirmation, which the post does not claim to have sought. The company also says it spent months trying to elicit a verified solve from other frontier models on some unsolved cipher and got none, but it does not say it ran this specific puzzle against competing models and watched them fail. And the author acknowledges steering the model away from the hardest-known ciphers, such as Kryptos K4, and toward one judged “unusually tractable,” which narrows the claim considerably: this was a curated task, not a blind stress test.
The gap the post does not address is whether Urquhart’s 1653 text, or prior scholarly discussion of the same cryptogram, was present in Fable 5.1’s training data. A solved answer that matches known historical facts about Urquhart is consistent with a genuine puzzle solve, but it is also consistent with a model recalling or reconstructing a previously published solution path. Treat this as one company’s report of one model completing one archival puzzle under conditions it selected, not as evidence that frontier models can now systematically clear historical cryptography’s backlog.
Reported by Vals AI on its company blog.
