An AI engineer used OpenAI’s GPT-6 Astra to decipher a coded letter from 1809 in six hours, AFP reports. The result offers a striking example of how general-purpose AI tools may help researchers tackle historical puzzles that have sat untouched for lack of time.
Watch Desk analysis
What happened
The letter, addressed to one of Napoleon’s generals, contains two dozen lines of symbols, numbers and letters describing French and allied troop positions. AFP reports that AI engineer Carter Church found it in an online academic archive and used GPT-6 Astra to crack the cipher. He consulted Japanese patent engineer Satoshi Tomokiyo, who maintains a list of unsolved ciphers, to check the solution.
Tomokiyo told AFP that most of the ciphers on his list have been cracked using AI since mid-September. That is his account of the list, not a measure of every historical cipher or of AI’s reliability across the field.
Why it matters
The useful change may be less about machines replacing expert code-breakers than about persistence and time. Church told AFP that conventional methods can work, but take considerable effort; many puzzles remain unattended while historians focus on other problems. A tool that can help sift through possibilities could make more of those sources worth investigating.
The same shift is appearing in other historical work. AFP describes AI-assisted analysis of medieval manuscripts and ancient scrolls, as well as transcription projects that can make large collections searchable. These are different tasks, but they share a practical payoff: researchers can ask questions of material that would be difficult to examine by hand at scale.
Our read
A cipher solved in an evening is a good story, and a useful prompt to ask what else has been waiting in the archive. It is not proof that AI can reliably interpret history on its own: Church sought a knowledgeable person’s check, and the report notes that some tougher ciphers may remain beyond current tools.
The exciting prospect is more human attention reaching neglected sources, not historians being shown the door. In this case, the machine did the patient searching; people still supplied context and a check on the answer.
What to watch
- Whether more solutions from unsolved-cipher lists are independently checked and documented.
- How researchers separate plausible AI-generated answers from historically sound interpretations.
- Whether AI-assisted transcription and analysis lead to findings that can be checked against the original sources.
Discussion spark: Should AI mainly be used to help researchers tackle neglected historical sources, or does the speed of these tools risk making verification the part people skip?
Sources and evidence
- 'Mixed feelings': AI cracks centuries-old ciphers (16 October 2026, 02:07 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.