Voynich Manuscript Latest Research: What We Know in 2026
The Voynich Manuscript Is Still Unread. Here’s What the Latest Research Actually Tells Us
Every so often, the Voynich Manuscript gets solved.
Or, more accurately, somebody announces that the Voynich Manuscript has been solved.
There has been no shortage of proposed languages, ciphers, hidden messages, elaborate hoaxes, secret medical texts and ingenious systems supposedly capable of making sense of the manuscript’s looping, stubbornly unreadable script.
And yet here we are.
The Voynich Manuscript is still undeciphered.
That is not because nothing is happening. Quite the opposite. Some of the most interesting recent research is moving away from the irresistible promise of “Here is what the Voynich Manuscript says” and toward a less spectacular but potentially more productive question:
What sort of system could have produced text like this in the first place?
Two pieces of research published in 2025 and 2026 offer intriguing answers to parts of that question. Neither translates the manuscript. Neither identifies its author. Neither finally tells us whether Voynichese is language, ciphertext, generated text, or something stranger.
What they do is make the mystery a little harder to hand-wave.
And after a century of modern attempts to crack the thing, that counts as progress.
First, what exactly is the Voynich Manuscript?
The manuscript is formally cataloged as Beinecke MS 408 at Yale University’s Beinecke Rare Book & Manuscript Library.
Yale describes it as a manuscript written in an unidentified script by an unknown author. Scientific examination has found its materials consistent with the fifteenth century, while Yale places its production in Central Europe around the end of the fifteenth or during the sixteenth century.
It surfaced on the modern rare-book market in 1912 through the Polish-American bookseller Wilfrid M. Voynich, whose name has been attached to it ever since.
If the writing were the manuscript’s only oddity, it would already be plenty.
It is not.
Its pages contain strange botanical drawings, circular astronomical and astrological diagrams, bathing figures, cosmological designs, apparent pharmaceutical material and pages that resemble recipes.
The pictures continually give the impression that meaning is just about to arrive.
It never quite does.
The plants look botanical, except when they do not. The diagrams look systematic, except that nobody has established what system they describe. The writing behaves enough like structured text to keep attracting linguists and cryptographers, but it has resisted attempts to establish what that structure means.
That tension is a large part of the manuscript’s peculiar grip on people.
Has the Voynich Manuscript been decoded?
No.
That answer is less exciting than the headlines that periodically circulate, but it is the accurate one.
Yale continues to describe the text as undeciphered and notes that cryptographic approaches have failed to decipher it, assuming it is encoded at all.
That last qualification matters.
There are several broad possibilities for Voynichese. It could encode a natural language. It could be ciphertext. It could represent some kind of constructed system. It could contain meaningless or semi-meaningful generated material.
Different researchers have spent decades finding evidence that appears to favor one possibility or complicate another.
The problem is that the Voynich Manuscript is very good at allowing an attractive idea to survive longer than it probably deserves.
Which brings us to 2025.
The 2025 Naibbe cipher: can a real cipher produce something that looks like Voynichese?
In 2025, researcher Michael A. Greshko published a paper in Cryptologia titled The Naibbe cipher: a substitution cipher that encrypts Latin and Italian as Voynich Manuscript-like ciphertext.
The experiment starts with an important distinction:
Greshko was not claiming to have found the Voynich Manuscript’s cipher.
Instead, he asked whether it was possible to construct a historically plausible cipher that could take ordinary Latin or Italian and produce ciphertext displaying many of the statistical peculiarities associated with Voynichese.
His resulting system, called the Naibbe cipher, is a verbose homophonic substitution cipher designed to be executable by hand using materials that could have existed in fifteenth-century Europe.
When used to encrypt a range of Latin and Italian texts, the resulting output reproduced a number of characteristics associated with the Voynich Manuscript while remaining decryptable with the cipher system.
That is interesting for a specific reason.
One objection to the idea that Voynichese could be ciphertext has been that ordinary substitution systems tend not to produce text with the manuscript’s peculiar statistical behavior.
Greshko’s work demonstrates that a more elaborate, but still hand-operable, substitution system can get substantially closer.
In other words, the existence of the Naibbe cipher shows that “ciphertext” cannot simply be dismissed because Voynichese looks statistically strange.
It does not establish that the Voynich Manuscript actually used this cipher.
Greshko also discusses significant ways in which his model fails to reproduce the real manuscript.
His conclusion is deliberately narrower: the ciphertext hypothesis remains viable, and a plausible substitution cipher would have to satisfy some unusually demanding structural constraints.
That is a much more interesting result than “Voynich solved.”
It gives other hypotheses something concrete to beat.
Then came another question: how constrained is Voynichese?
In 2026, Christophe Parisel released a computational preprint examining positional and directional structure in the manuscript.
And the word preprint matters here.
This is research made publicly available before the conventional peer-review and journal-publication process, so it should be treated as new evidence to examine, not as the final word on Voynichese.
Parisel’s analysis identified strong positional patterns in the manuscript’s graphemes and examined relationships both inside individual Voynich “words” and across the boundaries between them.
The study compared aspects of those patterns with English, French, Hebrew and Arabic, while explicitly warning that four comparison languages are nowhere near enough to make sweeping claims about natural language in general.
The study then tested two classes of text generator against a set of four positional signatures found in the manuscript.
One was a slot-based generator, in which Voynich-like words are assembled according to positional rules.
The other implemented a version of the Cardan-grille model associated with Gordon Rugg’s proposal that apparently complex Voynich text might have been mechanically generated without carrying ordinary linguistic meaning.
Neither tested generator class reproduced all four Voynich signatures simultaneously.
Again, the interesting part is what this does not mean.
It does not prove that Voynichese is encrypted.
It does not prove that it represents a natural language.
It does not eliminate every possible way of generating meaningless text.
Parisel explicitly states that the results do not rule out generator classes that were not tested.
What the work does provide is a more demanding benchmark.
If someone proposes a mechanism for producing Voynichese, that mechanism should increasingly be expected to reproduce not merely a few familiar-looking words or overall frequencies, but multiple interacting structural properties at once.
The goalposts have become harder to move.
That is useful.
So are these two studies pointing toward a cipher?
Possibly, but that is not yet the conclusion.
The 2025 Naibbe work demonstrates that a historically plausible hand cipher can reproduce many Voynich-like statistical properties.
The 2026 preprint argues that the manuscript contains positional constraints that the particular simpler structured generators it tested could not reproduce simultaneously.
Those findings are compatible with increased interest in cipher-like explanations.
They are not equivalent to evidence that one specific historical cipher produced the manuscript.
There is a larger methodological point here.
For years, Voynich research has been unusually vulnerable to explanations that can account for one striking property of the text.
A convincing theory has a harder job.
It has to account for many properties simultaneously.
It has to explain the structure of words, the distribution of glyphs, variations across sections of the manuscript, sequential relationships, the physical history of the object and whatever relationship, if any, exists between the writing and those frankly unhelpful plants.
The more measurable constraints researchers establish, the less room remains for a theory that merely produces something vaguely Voynich-looking.
The most important Voynich update may be that there is no dramatic update
This is perhaps the least internet-friendly conclusion available:
The Voynich Manuscript remains unsolved, and that is interesting.
We still do not know who wrote it.
We still cannot read it.
We still do not know with confidence whether Voynichese records language, encrypts language, generates language-like material, or does something for which our categories are inadequate.
But the question is becoming more precise.
A successful explanation now has more evidence to account for.
The Naibbe cipher provides an example of how unexpectedly elaborate ciphertext could be produced using historically plausible means.
Parisel’s analysis adds structural tests that proposed text-generation systems can be measured against.
Neither gives us the sentence everyone wants:
“Here is what the Voynich Manuscript says.”
Instead, they give us something research often produces before the satisfying answer arrives:
better ways to tell which answers are wrong.
For a manuscript that has survived this many attempted solutions, that may be exactly the kind of progress worth paying attention to.
And if somebody announces tomorrow morning that they have finally decoded the Voynich Manuscript?
We’ll have questions.
About the images
The manuscript shown with this article is Beinecke MS 408, the Voynich Manuscript, held by the Beinecke Rare Book & Manuscript Library at Yale University.
Courtesy of the Yale University Library.
Sources and further reading
Yale University Library: The Beinecke Cipher (Voynich) Manuscript
https://beinecke.library.yale.edu/beinecke/collections/beinecke-cipher-voynich-manuscript
Michael A. Greshko: The Naibbe cipher: a substitution cipher that encrypts Latin and Italian as Voynich Manuscript-like ciphertext, Cryptologia
https://www.tandfonline.com/doi/full/10.1080/01611194.2025.2566408
Christophe Parisel: Evidence of Layered Positional and Directional Constraints in the Voynich Manuscript: Implications for Cipher-Like Structure
https://arxiv.org/abs/2604.19762