In 1949, British solicitor T. E. Wood published an encrypted message consisting of just 21 letters:
FVAMI NTKFX XWATB OIZVV X
Wood revealed the encryption method he had used, but kept one crucial piece of information secret: the key. It was a passage from a book.
His reason for doing so was unusual, to say the least. Inspired by similar experiments conducted by the British parapsychologist Robert Thouless, Wood intended to communicate the secret key from beyond the grave after his death. Successful decryption would then provide evidence that the dead could communicate with the living.
Wood eventually died, but no correct key arrived from the afterlife. And the many codebreakers who tried to solve the 21-letter cryptogram without supernatural assistance were unsuccessful as well.
Until 2026.
AI finds the needle in the haystack
In the autumn of 2026, US-based computer expert Colin Reilly succeeded in determining the plaintext. [1] It reads:
HIER BIN ICH TOTSIENS TEW
“HIER BIN ICH” is German for “Here I am”, “TOTSIENS” is Afrikaans and can roughly be translated as “so long” or “see you”, while TEW represents Wood’s initials.
The mysterious key text turned out to be taken from the 1912 Luther Bible (Matthew 6:9–11). Finding this passage was a classic needle-in-a-haystack problem. With the assistance of AI, Reilly created a computer program that searched books that would have been available at the time and tested potential passages as keys. Eventually, it found the right one. Remarkably, Reilly had never done any codebreaking before.
An extraordinary season for codebreaking
Unlike Colin Reilly, I have been fascinated by encrypted messages from the pre-computer era for many years. I have written books, articles and academic papers about them, and there is no shortage of material. Tens of thousands of encrypted letters, newspaper advertisements, postcards, telegrams, diary entries, radio messages and handwritten notes survive in archives and private collections. And only a fraction of it has even been properly catalogued.
Many historical cryptograms can be solved quite easily with modern codebreaking techniques. Experts such as Israeli cryptanalyst George Lasry and German codebreaker Norbert Biermann have developed remarkable skills in this field.
But there are exceptions. About ten years ago, I compiled a list of what I regarded as the 50 most fascinating unsolved encrypted messages. Over the years, a few of them were deciphered.
Then came the autumn of 2026.
Within just a few weeks, a remarkable series of previously unsolved historical cryptograms were broken with the help of artificial intelligence. Several had been on my Top 50 list, including Wood’s message. Many of those who achieved these successes had previously been virtually unknown in the codebreaking community. What connected them was their use of AI tools.
Suddenly, tasks that had previously required extensive programming knowledge, specialized software and substantial manual research became easily accessible. AI can assist with writing analysis code, exploring large bodies of digitized material, testing hypotheses and evaluating vast numbers of potential solutions.
For that reason alone, the autumn of 2026 deserves a place in the history of codebreaking. Never before have so many notable historical cipher mysteries been solved within such a short period of time. And there is a good chance that we will never experience a period quite like it again.
AI does not make cryptographic mathematics disappear
This leads to much more important questions. Can AI break modern encryption, too? Could AI soon decrypt confidential messages? Could encrypted files become readable? Could an email protected with solutions such as Eviden GreenShield Mail suddenly be exposed?
The answer is no. The current wave of successful AI-assisted cryptanalysis concerns historical encryption systems from the pre-computer era. AI makes existing cryptanalytic approaches far easier to apply at scale. But this is fundamentally different from breaking modern cryptography.
Modern cryptography is based on mathematical problems and security margins that are fundamentally different from those of historical ciphers. Algorithms such as AES, RSA, Diffie-Hellman, SHA-2, Ascon, ML-KEM and ML-DSA have been studied extensively by researchers around the world.
AI can certainly help researchers look for weaknesses or explore possible attacks. What AI does not provide is a magical shortcut through the underlying mathematics. AI cannot simply make an infeasible cryptanalytic problem feasible because it is good at writing software or finding patterns.
In other words, the spectacular successes we are currently seeing against historical cryptograms should not be confused with a collapse of modern cryptography. The two worlds are fundamentally different.
So, by all means, let us enjoy the historical messages that AI is helping us read for the first time. AI may be opening locked doors from the past. That does not mean it holds the keys to today's cryptography.
Reference
[1] Klaus Schmeh, Wood’s cryptogram from the crypt: Another top 50 crypto mystery solved, October 4, 2026 (opens in a new tab)