Fable 5.1 Cracks the Cyphral Distich, a Cipher Unbroken Since 1655

A language model called Fable 5.1 has solved the Cyphral Distich, a two-line encrypted verse that has resisted decryption since it first appeared in 1655, according to a technical write-up published by the company Vals. The post drew 325 points and 118 comments on Hacker News, where readers argued over whether the solution reflects genuine cryptographic reasoning or a well-aimed pattern match. The answer, if it holds up, would close one of the longest-running open problems in historical cryptanalysis.

The distich is short — just two lines of verse — which is precisely what made it so stubborn. Short ciphertexts give analysts almost no statistical surface to work with: no repeated n-grams to exploit, no letter-frequency signal strong enough to separate real structure from noise. Generations of cryptographers treated it as a curiosity rather than a solvable target, and the few published attempts each proposed a different key without convincing the field.

According to Vals, Fable 5.1 approached the problem by generating and testing candidate readings against the constraints of the verse itself — meter, rhyme, and the plausibility of the resulting plaintext — rather than relying on a single decryption method. That iterative search over many hypotheses is where large language models differ most from classical attack tools, which tend to commit to one cipher model and fail cleanly when it is wrong. The company has not yet released the full plaintext and key in its post, which is the detail cryptographers will want before treating the result as settled.

Historical cipher claims have a mixed record, and the Hacker News thread reflected that. Commenters asked whether the model was given hints, whether the plaintext was verified against an independent source, and whether a shorter path — such as guessing a known phrase and reverse-engineering the key — could produce the same apparent success. Those are fair questions for any claim of this kind, and Vals will need to answer them with reproducible artifacts rather than a narrative of how the model got there.

The broader significance is less about one 17th-century puzzle than about what it says for the field. If a general-purpose model can grind through a short, underdetermined cipher by weighing linguistic plausibility, the same approach could be turned on damaged inscriptions, undeciphered scripts, and historical documents where the plaintext language is known but the encoding is not. That is a different use case from the benchmark tasks AI labs usually report, and it maps onto real work that humanities researchers have struggled with for decades.

What to watch next is verification. Independent cryptographers will want the full key, the plaintext, and a clear statement of what information the model had access to before it started. If the solution survives that scrutiny, the Cyphral Distich moves from unsolved to solved — and Fable 5.1 becomes a data point in the argument over whether language models can do genuine cryptanalysis or only imitate its results.