How AI is helping historians read secrets locked away for centuries

Artificial intelligence is helping historians and cryptologists decipher thousands of encrypted historical documents, revealing long-hidden information about political intrigues, medical practices and personal correspondence preserved in archives and libraries around the world.

Researchers estimate that roughly 1% of all archival material is fully or partially encrypted, leaving a vast body of historical records inaccessible without specialized decoding techniques. These documents contain information ranging from diplomatic intelligence and military affairs to medical remedies and private relationships, says the BBC.

Recent advances in artificial intelligence and machine learning have led to several notable breakthroughs.

Among them is the decoding of the Borg Cipher, a 400-year-old manuscript held in the Vatican library. Researchers found the text contained unusual medical treatments, including recommendations to use high-quality red wine or fermented nutmeg to treat dysentery. Historians believe such remedies may have been concealed to avoid accusations of witchcraft.

AI-assisted analysis has also helped decode letters written by Mary Queen of Scots, revealing her involvement in efforts to regain her throne and shedding light on her strained relationship with her son, King James.

In another case, researchers deciphered a 500-year-old letter written by Holy Roman Emperor Charles V that revealed concerns about a possible assassination plot involving an Italian mercenary. A separate breakthrough uncovered the contents of a 1637 letter from Sigismund Heusner von Wandersleben during the Thirty Years' War, exposing warnings about conspiracies among Sweden's Protestant allies.

Until recently, deciphering such documents was often a slow and labor-intensive process. Historians and cryptologists first had to convert handwritten symbols into digital text, a task complicated by faded ink, damaged pages and difficult handwriting. Researchers say transcribing a two-page letter can take an entire day.

The challenge extends beyond transcription. While some historical ciphers rely on simple letter substitutions, others employ decoy symbols or multiple symbols representing the same letter to obscure their meaning. In some cases, researchers do not even know the language underlying the encrypted text.

New AI-powered tools are helping accelerate the process.

Platforms such as Transkribus and systems being developed through the Descrypt project use machine learning to recognize historical handwriting and scripts from different centuries, converting manuscripts into machine-readable text.

Researchers are also developing systems that combine transcription and decryption into a single process. Using large language models alongside image-recognition technology, these tools can identify patterns in encrypted manuscripts and generate possible translations.

The technology has proven particularly useful for testing large numbers of cipher keys rapidly. Researchers reported that an AI chatbot decoded a 500-symbol extract of the Borg Cipher in 29 minutes, significantly reducing the time required for analysis.

Experts say the long-term objective is to develop tools capable of assisting with a broad range of historical puzzles, including scripts that remain undeciphered.

Future applications could include efforts to interpret ancient writing systems such as Linear A, a script dating back about 4,000 years, and the symbols found on the Phaistos Disc. By building databases of known cipher systems and historical scripts, researchers hope to train AI models to recognize and analyze forms of writing that have so far resisted traditional methods.

While historians caution that human expertise remains essential for verifying interpretations and providing historical context, researchers say artificial intelligence is rapidly becoming a powerful tool for unlocking information that has remained hidden for centuries.