Anthropic's new model successfully deciphered a historical cryptogram. The Claude Fable 5.1 system solved the cipher of writer Thomas Urquhart. The puzzle, known as the Cyphral Distich, had remained unsolved for three hundred and seventy years. This achievement was reported by specialists from the well-known Vals AI project. The cryptogram consists of two lines of thirty-two numbers each. Experts discovered the cipher at the end of the treatise Logopandecteision, published back in 1653. In 1899, the cipher was brought to public discussion and subsequently appeared in specialized twentieth-century literature. Researcher Klaus Schmeh ranked the cipher twenty-eighth on a famous list of the fifty most well-known undeciphered messages. All previous decryption attempts relied on frequency analysis, and none of them yielded a positive result.
The model drew attention to two important features of the text. The cipher is printed immediately after thirty-two numbered wishes by the author. The second feature concerns a special accompanying poem in the book. From these facts, the decryption rule for the historical cipher can be reconstructed. Each number refers to a specific procuration within the book's text, indicating the ordinal number of a word in that section. The first letters of these words combine to form the final message. The deciphered text turned out to be a couplet in support of King Charles. Urquhart was a well-known supporter of the Stuart dynasty of his time. The model also analyzed an eight-line stanza consisting of 285 numbers — a cryptogram encoded in the 1652 treatise The Jewel. Nine letters in the fifth line could not be deciphered.
The model worked on the complex task for forty-four minutes, consuming one hundred and seventy-six thousand computational tokens. The experiment's author, Gaby Jaffe, set a general goal for the model, having spent several months attempting to obtain a solution from other systems. The model's key advantage proved to be its ability to select tasks within its capabilities. Historical cipher specialists have not officially confirmed the decryption. A report from Reticuli Labs appeared on GitHub, in which the authors fully disputed the results of the Vals AI project. The first objection concerns the decryption algorithm itself: the letter K should appear in the eleventh position of the first line, yet the corresponding section contains eighty words.
Researchers identified ten such critical discrepancies in the text. The lab tested sixty-five variants of letter counting, with the best result yielding eight matches out of sixty-four letters — a figure consistent with ordinary statistical chance. The second objection concerns the very origin of the historical cipher: the digitized copy of the 1653 edition contains no numerical lines. Both cryptograms are known only from Urquhart's 1899 biography. The parties disagreed over the historical sources used: the Vals AI project cited the 1834 collected works. Separately, Levent Alpöge previously found a counterexample to the Jacobian conjecture, a central problem in modern algebraic geometry.