The world of cryptography has witnessed a historic breakthrough: Anthropic's Claude Fable 5.1 model has successfully deciphered the famous Cyphral Distich cryptogram, composed by Scottish writer Thomas Urquhart more than 370 years ago. My colleagues from the Vals AI research project have confirmed this result, and I must admit that this event has the potential to overturn perceptions of artificial intelligence's capabilities in historical analysis.
The Cyphral Distich consists of two lines, each containing 32 numbers. This cryptogram was discovered at the end of Urquhart's treatise Logopandecteision, published in 1653. For centuries, it remained one of the most intriguing unsolved riddles: in 1899, the cipher was brought up for public discussion in Notes and Queries, and in the 20th century it firmly entered cryptographic literature.
Leading historical cipher researcher Klaus Schmeh even placed this cryptogram at 28th in his ranking of the 50 most famous undeciphered messages. All previous attempts at cracking it were based on frequency analysis and substitution methods, but none succeeded — and here is why.
The key turned out to be inside the book
The Fable 5.1 model discovered what cryptographers had missed for centuries. The cipher is printed immediately after Urquhart's 32 "procuritations" — numbered wishes ending with phrases like "this is the wish" or "hope." The author specifically emphasizes the number 32 in the text. An accompanying poem promises the reader that they will find their own wishes and the author's thought within the cipher.
The rule turned out to be elegantly simple: the i-th number in a line refers to the i-th procuritation, the value indicates the ordinal position of a word within it, and the first letters of those words form the message. The decrypted text is a couplet in support of Charles II: "God support King Charles the Second and make him supreme ruler of this land."
The result is self-verifying: each line contains exactly 32 letters, the lines rhyme and form a couplet, just as the author promised. Urquhart's political stance as a supporter of the Stuarts fully aligns with the content.
A second cryptogram and doubts
Using the same method, the model deciphered the Cyphral Octastich — an eight-line poem of 285 numbers from the 1652 treatise The Jewel. Here, the numbers refer to pages of the book. However, nine letters in the fifth line remained undeciphered — confirming them requires a physical copy of the edition.
Of course, as expected, the result sparked debate within the professional community. Specialists from Reticuli Labs presented a report challenging Vals AI's conclusions. They point to inconsistencies in the algorithm: to obtain the word KING, the letter "K" must be at the 11th position, but in the corresponding procuritation of 80 words, none begin with that letter. There are ten such discrepancies in total. Alternative counting methods yielded only eight matches out of 64 letters, which corresponds to the level of randomness.
Furthermore, critics note: in the digitized copy of the 1653 edition, there are no numerical lines after the procuritations at all — both cryptograms are known from Urquhart's biography published in 1899.
My analysis: Regardless of the outcome of the dispute, the very fact that an LLM is capable of formulating and testing complex cryptographic hypotheses, relying on contextual clues in historical documents, is a colossal step forward. Even if the specific decipherment turns out to be debatable, the methodology demonstrated by Fable 5.1 opens new horizons for applying AI in historical cryptography. The controversy surrounding the result only underscores the importance of verification and healthy skepticism in our field.