Anthropic's newest language model, Claude Fable 5.1, has accomplished what cryptographers failed to do for over three and a half centuries: it found the key to the Cyphral Distich cipher left by Scottish writer Thomas Urquhart. This cryptogram, consisting of two lines of 32 numbers each, was published in the treatise Logopandecteision in 1653 and has since been considered one of the most complex unsolved puzzles in history.
Until recently, all decryption attempts were based on frequency analysis and the assumption that the key had to be sought outside the text of the work itself. Historical cipher researcher Klaus Schmeh even placed this riddle at 28th place in his ranking of the 50 most famous undeciphered messages. However, as it turned out, the solution lay on the surface—in the structure of the book itself.
The Key Is in the Book Itself
My analysis shows that the genius of Fable 5.1's solution lies in rejecting standard approaches. The model noticed that the cipher was printed immediately after 32 numbered "procurations"—the author's wishes—and that Urquhart specifically highlights the number 32 in the text. Cross-referencing this with the accompanying poem, which promises the reader that they will find their own wishes in the cipher, the AI derived an elegant rule: the i-th number in a line refers to the i-th procuration, the value indicates the ordinal number of a word, and the first letters of those words form the message.
The decrypted text turned out to be a couplet in support of Charles II: "God support King Charles the Second and make him supreme ruler of this land." This fully aligns with Urquhart's political views, as he was a staunch supporter of the Stuarts.
The Second Cryptogram and Skeptics
Using the same method, the model also decrypted the more complex Cyphral Octastich—an eight-line poem of 285 numbers from another Urquhart treatise. Here, the numbers referred to pages of the book, and only nine letters remained unconfirmed due to the lack of a fully digitized version of the edition.
However, the triumphant announcement by Vals AI, who conducted the experiment, has already been challenged by the professional community. Specialists at Reticuli Labs conducted an independent replication and found critical inconsistencies. The main objection: to obtain the word KING in the decryption, a certain word in the corresponding procuration must begin with the letter "K," but in that section of 80 words, no such word exists. In total, they counted ten such mismatches, and when testing 65 counting variants, the best result was only eight matches out of 64 letters—which corresponds to the level of randomness.
Moreover, questions have also arisen about the source of the cipher: the original 1653 edition contains no numerical lines at all—they are known only from an 1899 biography of Urquhart, whereas Vals AI referred to an 1834 collection of works.
My verdict: although Fable 5.1's breakthrough is impressive in terms of methodology and unconventional thinking, it is premature to speak of a complete victory of AI over historical cryptography. The situation reminds me of the classic problem of confirmation in science: a beautiful hypothesis requires ironclad evidence. Until independent experts gain access to the primary sources and confirm the algorithm on other examples, this case should be viewed as an intriguing experiment rather than a definitive solution. Nevertheless, the model's very ability to select a manageable task from a multitude of unsolved ones and not retreat in the face of difficulties is an important step forward in the development of AI.