Breaking Historical Ciphertexts with Modern Means

Klaus Schmeh and Elonka Dunin open by noting that many thousands of old ciphertexts survive in archives and collections, and that a large number remain unsolved. Historical codebreaking differs from modern cryptanalysis: the algorithm is usually unknown, and the aim is recovering the plaintext rather than a key. The talk starts with monoalphabetic substitution, where frequency analysis is the standard first step, illustrated with a 17th-century alchemical cryptogram and a 19th-century Freemason document.

Short texts defeat frequency analysis, so the presenters turn to hill climbing: a random substitution key is altered, and the change is kept only if a letter-frequency score improves. Simulated annealing helps the search escape local maxima. For transposition ciphers such as turning grilles, scoring relies on letter pairs instead, and a large grille challenge published online was solved within a few hours. The nomenclator section covers codes that mix letters with whole words, and the Zodiac Killer's Z340 cipher, which was solved by a three-person team in 2020 using computers.

The closing section covers unsolved examples including the Voynich manuscript and the CIA's Kryptos part four. The presenters say generative AI has not yet cracked them. In the Q&A they addressed layered encryption, one-time pads and practice resources. The takeaway is that historical codebreaking is an active research area, with hill climbing currently the strongest general method, and many puzzles remain open for anyone who wants to try them.