A key breakthrough in historical cryptography: Anthropic's new Claude Fable 5.1 model has managed to decipher the famous Cyphral Distich cryptogram, created by Scottish writer Thomas Urquhart in 1653. This puzzle remained unsolved for more than three and a half centuries and was among the top 50 most famous unsolved ciphers in the world.
The key to the solution lies in the structure of the book itself
The cryptogram consists of two lines of 32 numbers, printed at the end of the treatise Logopandecteision. Previous researchers unsuccessfully applied frequency analysis and various substitution methods, assuming that the key had to be sought outside the text. However, Fable 5.1 drew attention to a fundamentally different approach: the numbers refer to 32 "procuritations"—the author's numbered wishes printed earlier in the same book.
The decryption algorithm turned out to be elegantly simple: the i-th number in a line points to the i-th procuritation, and its value determines the ordinal position of a word within it. By collecting the first letters of these words, the model obtained a couplet in support of King Charles II: "God support King Charles the Second and make him the supreme ruler of this land". The result fully aligns with Urquhart's political views, as he was a staunch supporter of the Stuart dynasty.
The second cryptogram and skeptics' doubts
Using the same method, the model also deciphered the more complex Cyphral Octastich—an eight-line verse of 285 numbers hidden in another treatise by Urquhart. Here, the numbers referred to pages of the book, but nine letters in the fifth line remained undeciphered due to a missing leaf in the digitized copies.
The methodology of the experiment is also noteworthy: the model worked autonomously for 44 minutes, consuming 176,000 tokens and with virtually no operator hints. The study's author, Gebi Jaff, emphasizes that Fable 5.1 demonstrated a unique ability to select a feasible task from a multitude of unsolved ones and not retreat after initial failures—a quality lacking in other advanced LLMs.
However, the scientific community received the result with mixed feelings. Independent experts from Reticuli Labs contested the decryption, pointing out ten positional mismatches: in the corresponding procuritations, words beginning with the required letters are absent. Their own enumeration of 65 counting variants yielded only eight matches out of 64 letters—consistent with the level of randomness. Moreover, critics note that the first printed edition of 1653 contains no numerical lines at all, and both cryptograms are known only from Urquhart's biography published in 1899.
My expert assessment: This situation demonstrates the dual nature of AI discoveries. On one hand, the model's ability for unconventional thinking is impressive. On the other, the lack of a reproducible result and discrepancies in sources remind us that even the most elegant solutions require rigorous verification. In the world of cryptography, as in the world of digital assets, trust is built on evidence, not on loud claims.