Kryptos K4: Every Cipher It Could Still Be
Ninety-seven letters on a copper screen at CIA headquarters have gone unread in public since 1990. Their author has given away twenty-four of them, and those twenty-four are enough to rule out most of the ways the passage could have been made. This page is about what is left: nineteen cipher families and two ways of combining them, each running here in the browser on any text and any key, in both directions, and each one click from being pointed at K4.
The sculpture
Kryptos is a sculpture by Jim Sanborn in a courtyard of the CIA's New Headquarters Building in Langley, Virginia, dedicated on 3 November 1990.[1] Its main element is a curved copper screen with letters cut clean through it. The left half is ciphertext. The right half is a table of shifted alphabets, a Vigenère tableau, built on an alphabet that starts with the letters of the word KRYPTOS and carries on with the rest in order.
The ciphertext runs to 865 letters and four question marks, in four passages that everyone now calls K1 to K4.[2] Sanborn is a sculptor. The systems came from Ed Scheidt, a former chairman of the CIA's Cryptographic Center, who put it this way: “I set the codes out for him, which means he didn't have to know the mathematics of it.”[3]
Three passages, solved three times
The first three passages were broken three times over, by people who did not know about one another. A team at the NSA took up a challenge from the CIA's deputy director in late 1992 and reported all three solved in a 1993 memo, in private; the paperwork came out under a FOIA request in 2013.[4] David Stein, at the CIA, had his breakthrough on 21 February 1998.[5] Jim Gillogly did it by computer and was the first to say so publicly, in June 1999.[6]
EMUFPHZLRFAXYUSDJKZLDKRNSHGNFIVJYQTQUXQBQVYUVLLTREVJYQTMKYRDMFD
Between subtle shading and the absence of light lies the nuance of iqlusion.
VFPJUDEEHZWETZYVGWHKKQETGFQJNCEGGWHKK?DQMCPFQZDQMMIAGPFXHQRLGTIMVMZJANQLVKQEDAGDVFRPJUNGEUNA…
It was totally invisible. How's that possible? They used the Earth's magnetic field. X The information was gathered and transmitted undergruund to an unknown location. X Does Langley know about this? They should: it's buried out there somewhere. X Who knows the exact location? Only WW. This was his last message. X Thirty-eight degrees fifty-seven minutes six point five seconds north, seventy-seven degrees eight minutes forty-four seconds west. X Layer two.
ENDYAHROHNLSRHEOCPTEOIBIDYSHNAIACHTNREYULDSLLSLLNOHSNOSMRWXMNETPRNGATIHNRARPESLNNELEBLPIIACAE…
Slowly, desparatly slowly, the remains of passage debris that encumbered the lower part of the doorway was removed. With trembling hands I made a tiny breach in the upper left-hand corner. And then, widening the hole a little, I inserted the candle and peered in. The hot air escaping from the chamber caused the flame to flicker, but presently details of the room within emerged from the mist. X Can you see anything? Q
K1 and K2 are one system with two keywords. Take the KRYPTOS alphabet, use it for the plaintext and for the ciphertext, and shift it by a word that repeats: PALIMPSEST for K1, ABSCISSA for K2. The American Cryptogram Association calls this a Quagmire III.[7] It is the operation the tableau on the right half of the screen exists to perform, and both passages run in the workbench below exactly as they do on the copper.
K3 is a different kind of thing: a transposition. Every letter of the plaintext is there, in a different order. The text is adapted from Howard Carter's account of opening Tutankhamun's tomb.[6]
Each plaintext has a misspelling in it: IQLUSION, UNDERGRUUND, DESPARATLY. Sanborn has said the first two are deliberate.[8] K2 also has an accident. In 2006 it emerged that a letter was missing from the copper: restore an S near the end and the last words change from ID BY ROWS to X LAYER TWO.[9] The K2 example below includes that S. It matters for K4 too, as a reminder that what was carved is not guaranteed to be what was intended.
Ninety-seven letters, twenty-four known
K4 is the last 97 letters on the screen. This is how it sits there: four letters at the end of a line, after a question mark, then three full lines of 31.
The marked cells are the 24 letters of plaintext Sanborn has released, each under the ciphertext letter it stands for. Positions count from 1 at the O of OBKR.
Sanborn has given four clues. In November 2010: positions 64 to 69 decipher to BERLIN. In November 2014: the next five are CLOCK. In January 2020: positions 26 to 34 are NORTHEAST. In August 2020: the four before that are EAST.[10][11] That makes EASTNORTHEAST at 22–34 and BERLINCLOCK at 64–74. Cryptanalysts call known plaintext like this a crib.
Twenty-four letters out of 97 does not sound like much. They say a good deal before any cipher is assumed:
- Letters stay in place. Sanborn has described BERLINCLOCK and NYPVTTMZFPK as “a one-to-one match”.[2] Whatever else happened, a shuffle like K3's was not the last step.
- There is more than one alphabet at work. E becomes F, then G, then Y. T becomes V, R and S. Going the other way, the ciphertext letter P stands for R twice and for C once.
- A letter can encipher to itself. S does at position 33 and K does at 74. That removes every system that can never do so, as Enigma could not.
- R becomes P twice, 38 places apart, at positions 28 and 66. In any system where the key picks a row of a fixed tableau, the key at those two positions is the same.
- The ciphertext is flat. English has an index of coincidence near 0.067, random letters 0.038, and K4 comes in at 0.036. Scheidt has said that in the fourth part he deliberately masked the letter frequencies that make English breakable.[12]
Found, but not solved
In August 2025 Sanborn announced that he would auction the solution.[11] Before the sale, two writers found it. On 2 September Richard Byrne, at Jarett Kobek's request, went through papers Sanborn had given to the Smithsonian's Archives of American Art, and the plaintext of K4 was among them.[13][14] They told Sanborn the next day. They have not published it, and they have been exact about what it is. Kobek: “Rich and I recovered the plaintext. There's no way on earth that this is a cryptographic solve, and we have not claimed that.”[14] Sanborn asked the Smithsonian to seal the files for fifty years, and it did.[13]
The auction closed on 20 November 2025, widened from the K4 solution to Sanborn's whole Kryptos archive, at $962,500.[15] In June 2026 the investment firm Paradigm announced itself as “the new stewards of the Kryptos secret” and put up a site that checks a submitted answer against a keyed hash of the real one.[16]
About a week before the sale closed Sanborn published a last set of clues: that a 1986 trip to Egypt and the fall of the Berlin Wall both figure in the solution, and that the BERLINCLOCK of the earlier clues is the World Clock in Berlin. He also described a second passage of 97 characters, now called K5, made with a similar system and with BERLINCLOCK in the same position.[17] Paradigm says it plans to release K5. As of November 2026 it has not.[16]
So the position, as I write this, is an odd one. Several people know what K4 says. Nobody has published how it was enciphered, and that question is the one this page is about.
One operation, many keys
Nearly every candidate is a variation on a single operation. Number the letters of a plain alphabet 0 to 25. To encipher a letter, take its number x, combine it with a key value k, and read the result y off a cipher alphabet:
The two alphabets can be anything: A to Z, the KRYPTOS alphabet, a different keyword on each side. What separates one family from the next is only where k comes from at each position.
- A word that repeats. Quagmire, with Vigenère as its plainest case.
- Two words of different lengths, added. Summed keys.
- A text as long as the message, or letters drawn at random. Running key and the one-time pad.
- The message itself, a few letters back. Autokey.
- A stream of digits grown from a short primer by adding pairs. Gromark.
- Where the letter sits on the copper: its column plus its row, or a key laid along some route through the grid. The workbench draws both.
- A repeating word that keeps getting interrupted.
The rest are the systems that picture does not cover: a carry that passes from letter to letter, a tableau whose rows are unrelated, pairs of letters enciphered together (Digrafid, Hill, a free table of digraphs), and machines whose alphabets move as they work (Chaocipher, the M-209).
What the known letters settle
For any system of the first kind, the 24 known letters turn into 24 known key values the moment the two alphabets are fixed: at each known position x and y are both in hand, so k is too. The workbench shows this as “the key the known letters ask for”. A candidate then lives or dies on whether its key generator can produce those values at those places. A repeating key of period 7 has to show the same value at positions 22 and 29. A Gromark chain has to obey its own recurrence across both clues. Most candidates cannot.
I worked through the families one at a time, and the method was the same for each. Fix what kind of system it is, then ask whether any key of that kind turns the known plaintext into the ciphertext at those positions. For alphabets built from keywords that is an enumeration: about 1.5 million alphabets, in five pairings of plain and cipher alphabet, at every period up to 52. For alphabets that are free permutations it becomes a constraint problem for a SAT solver, which either finds a consistent key or proves there is none. Where neither settles it, as with Gromark and its two unknown alphabets, a simulated annealer looks for alphabets that make the rest of the text read as English. The searches were also run on planted messages, to show that each finds a key when one is there.
| Family | Standing | Where the known letters leave it |
|---|---|---|
| Narrowed | Open only with non-keyword alphabets: Quagmire III in Beaufort form at period 23 or 26, Quagmire IV at period 8, 13, 16, 19, 20, 23, 24 or 26, and any type above period 26; every other period to 26 is impossible. | |
| Narrowed | Open when the periods add up to 26 or more (the cribs cannot reach it); shorter pairs are dead. | |
| Narrowed | Dead for keyword alphabets to period 12; open with arbitrary alphabets or longer periods. | |
| Narrowed | Dead for keyword alphabets at periods 1 to 23; open with arbitrary alphabets. | |
| Out of reach | Open: a key text nobody has identified, or a one-time pad, cannot be tested from 97 letters. | |
| Narrowed | Open with arbitrary alphabets at longer lags: plaintext-keyed Beaufort form from lag 5, additive from lag 11; ciphertext-keyed is impossible at lags 1 to 12. | |
| Closed | Dead as defined by the ACA (any keyed alphabet); kept because the open autokey cells are its relatives. | |
| Narrowed | Dead for keyword alphabets (word, letter and free-phase interruptions); open with arbitrary alphabets. | |
| Open | Open pending batch runs: 6-digit primers, keyword primers, letter chains (base 26) and one-slip streams; 2- to 5-digit primers with clean chains are dead. | |
| Closed | Dead for every dictionary keyword in 576 conventions, both as written and stripped of repeated letters as the ACA rule says; kept as the reference for the ACA construction. | |
| Narrowed | Dead for keyword alphabets on every layout with enough crib constraints; open with arbitrary alphabets. | |
| Narrowed | Dead for keyword alphabets on the 13,364 routes with enough crib constraints; open with arbitrary alphabets. | |
| Out of reach | Open and unfalsifiable from 97 letters once there are 8 or more independent alphabets. | |
| Out of reach | Open: with unknown strip alphabets there are 88 bits of key per strip and 97 letters cannot fix them. | |
| Narrowed | Impossible for any pair of alphabets at 2 to 7 digraphs per group; open at 16, 24, 25, 26, 29, 30, 35 (K4 with its '?' in front) and 8, 14, 22, 24, 31, 46 (K4 plus one lost character). | |
| Narrowed | Dead for 2x2 to 4x4 with A=0..Z=25 numbering, alone or under a periodic key; open with an arbitrary numbering of the letters. | |
| Out of reach | Open: a free table of 676 digraphs is beyond anything 24 known letters can exclude. | |
| Out of reach | Open: two alphabets that reshuffle after every letter leave nothing for 24 cribs to test. | |
| Out of reach | Open: six pin wheels and 27 lug bars are a key of about 170 bits, out of reach of 24 cribs. |
2 closed, 10 narrowed, 1 open and 6 out of reach. “Dead” means no key of that kind fits the 24 letters. “Keyword alphabets” are the ones a word can generate; “arbitrary” ones are any of the 26! orderings. Click a family to open it in the workbench.
Three things are missing from the list on purpose. Trifid, Ragbaby and the Wheatstone cryptograph are closed for every key, so there is nothing to try. Transposition is out because the clues tie each plaintext letter to its own position. And a hand-drawn chart that is not a substitution table cannot be written down as a family at all; one that is belongs under the multi-alphabet tableau.
None of this says what K4 is. “Closed” means closed under the stated assumptions, and the largest of those is that the clues are right and the text was enciphered without a slip. K2 lost a letter on its way to the copper. Scheidt is quoted as saying he would consider K4 to be more than one stage,[2] and a mask over a cipher is exactly what the Compose combinator below is for. Keith Adler has published an independent map of the same ground.[18]
The workbench
Pick a family, set its key, and type in either box: plaintext on the left enciphers, ciphertext on the right deciphers. Under the boxes are the cipher letter by letter, the key written out in full, and what that key makes of K4. The worked examples are published ones wherever a published one exists, and the page says so when its output matches letter for letter. Everything runs in your browser; nothing you type is sent anywhere.
Quagmire I–IV
NarrowedA keyword that repeats, on alphabets that may themselves be keyed. Vigenère, Beaufort, Variant and Gronsfeld are the straight-alphabet case; K1 and K2 are type III on the KRYPTOS alphabet.
Against K4Open only with non-keyword alphabets: Quagmire III in Beaufort form at period 23 or 26, Quagmire IV at period 8, 13, 16, 19, 20, 23, 24 or 26, and any type above period 26; every other period to 26 is impossible.
Key
Straight: plain A–Z on both sides. I: keyed plain alphabet. II: keyed cipher alphabet. III: one keyed alphabet for both. IV: a different keyed alphabet for each.
KRYPTOSABCDEFGHIJLMNQUVWXZ
A word, or shifts written 3,1,4 or 31415.
The plaintext letter the key letters stand under. The ACA convention is A; Kryptos used K.
Vigenère adds the key, Variant subtracts it, Beaufort subtracts the letter from the key.
✓ Letter for letter what the sculpture gives.
Letter by letter
| # | 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| plain | B | E | T | W | E | E | N | S | U | B | T | L | E | S | H | A | D | I | N | G | A | N | D | T | H | E | A | B | S | E | N | C | E | O | F | L | I | G | H | T | L | I | E | S | T | H | E | N | U | A | N | C | E | O | F | I | Q | L | U | S | I | O | N |
| key | P | A | L | I | M | P | S | E | S | T | P | A | L | I | M | P | S | E | S | T | P | A | L | I | M | P | S | E | S | T | P | A | L | I | M | P | S | E | S | T | P | A | L | I | M | P | S | E | S | T | P | A | L | I | M | P | S | E | S | T | P | A | L |
| cipher | E | M | U | F | P | H | Z | L | R | F | A | X | Y | U | S | D | J | K | Z | L | D | K | R | N | S | H | G | N | F | I | V | J | Y | Q | T | Q | U | X | Q | B | Q | V | Y | U | V | L | L | T | R | E | V | J | Y | Q | T | M | K | Y | R | D | M | F | D |
The key, written out
- Plain alphabet
- KRYPTOSABCDEFGHIJLMNQUVWXZ
- Cipher alphabet
- KRYPTOSABCDEFGHIJLMNQUVWXZ
- Form
- Vigenère
- Key
- PALIMPSEST
- Indicator
- K
- Period
- 10
- Tableau: each key letter’s row, under the plain alphabet
K R Y P T O S A B C D E F G H I J L M N Q U V W X Z P P T O S A B C D E F G H I J L M N Q U V W X Z K R Y A A B C D E F G H I J L M N Q U V W X Z K R Y P T O S L L M N Q U V W X Z K R Y P T O S A B C D E F G H I J I I J L M N Q U V W X Z K R Y P T O S A B C D E F G H M M N Q U V W X Z K R Y P T O S A B C D E F G H I J L P P T O S A B C D E F G H I J L M N Q U V W X Z K R Y S S A B C D E F G H I J L M N Q U V W X Z K R Y P T O E E F G H I J L M N Q U V W X Z K R Y P T O S A B C D S S A B C D E F G H I J L M N Q U V W X Z K R Y P T O T T O S A B C D E F G H I J L M N Q U V W X Z K R Y P
This key against K4
2 of 24 known letters right
22–34 reads OKFTLHMUIDNIL for EASTNORTHEAST · 64–74 reads TDKJNXHVOFE for BERLINCLOCK
The key the known letters ask for
With these alphabets and this form, the key letter that turns each known plaintext letter into the ciphertext letter under it. Whatever generates the key has to produce exactly this at these 24 places, so a repeating key would show up here as a repeat.
| # | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| plain | E | A | S | T | N | O | R | T | H | E | A | S | T |
| cipher | F | L | R | V | Q | Q | P | R | N | G | K | S | S |
| key | R | D | U | M | R | I | Y | W | O | Y | N | K | Y |
| value | 1 | 10 | 21 | 18 | 1 | 15 | 2 | 23 | 5 | 2 | 19 | 0 | 2 |
| # | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| plain | B | E | R | L | I | N | C | L | O | C | K |
| cipher | N | Y | P | V | T | T | M | Z | F | P | K |
| key | E | L | Y | O | I | E | C | B | A | Q | K |
| value | 11 | 17 | 2 | 5 | 15 | 11 | 9 | 8 | 7 | 20 | 0 |
Alphabets are written the same way everywhere: A–Z, KRYPTOS, a keyword, a keyword mixed the ACA's columnar way, a shuffle from a seed, or 26 letters typed out. A seed names the same alphabet on every machine.
How it is checked
A cipher that is implemented wrongly still produces letters, and they look as much like ciphertext as the right ones do. So nothing here is trusted on sight.
- Published examples, letter for letter. K1 and K2 from the sculpture; the ACA's own worked examples for Quagmire I to IV, Running Key, Interrupted Key, Condi, Gromark, Periodic Gromark and Digrafid at two group sizes;[7] the worked example in Moshe Rubin's disclosure of the Chaocipher algorithm;[19] the 26-letter check of a published M-209 key list. Each one is a preset above.
- A reference implementation. The engine in this page is a TypeScript port of the Python package the search ended with. The port is held to the Python on 542 keys across every family: each ciphertext, each decryption, each resolved key and each refusal, some 3,500 comparisons in all.
- The same thing computed twice. Addition with carry has to agree with big-integer arithmetic. Two Vigenère passes have to equal one summed key. A Quagmire has to equal a tableau of its own rows. The key the known letters ask for, typed back in as a key, has to reproduce both clues.
The gaps are worth stating. The strip cipher, the digraph table, summed keys, the grid and route keys, the affine key, addition with carry and the multi-alphabet tableau have no published example behind them, only round trips and a second derivation. And an interrupted key whose trigger is the very letter being produced has no unique inverse: the page finds the first consistent reading and tells you how many letters could have gone the other way.
Written in November 2026. The account of 2025 and 2026 will date; the sources below are where to look for what has changed.
- Central Intelligence Agency, “About the Kryptos Sculpture”.
- Richard Bean, “Cryptodiagnosis of ‘Kryptos K4’”, HistoCrypt 2021. The letter count is Bean's; the remarks by Sanborn and Scheidt are quoted there from the Kryptos discussion group.
- NOVA scienceNOW, 24 July 2007, PBS, transcript.
- Elonka Dunin, “Elonka's FOIA request”, 2013: the NSA's Kryptos documents and what they show.
- David Stein, “Cracking the Courtyard Crypto”, Studies in Intelligence 43(1), 1999, as released to the National Security Archive.
- Jim Gillogly, post of 16 June 1999, mirrored by Elonka Dunin.
- American Cryptogram Association, cipher types. Each type's page carries the worked example reproduced here.
- Elonka Dunin, Kryptos FAQ.
- Kryptos Group, “Kryptos Part 2 Was Wrong”, 19 April 2006.
- Klaus Schmeh, “Jim Sanborn publishes another Kryptos clue”, 2020, which lists all four clues with their positions.
- Elonka Dunin, Kryptos page and timeline.
- “CIA Experts Still Spooked by Kryptos Puzzle”, NPR, 2005. The wording is the reporter's paraphrase of Scheidt.
- “A Solution to the CIA's Kryptos Code Is Found after 35 Years”, Scientific American, October 2025.
- “Kryptos K4 Uncovered”, an interview with Jarett Kobek and Richard Byrne, Zona Motel, October 2025.
- RR Auction, “Jim Sanborn's Complete Kryptos Archive Sells for $962,500 at Auction”, November 2025.
- Paradigm, “Kryptos”, 12 June 2026.
- “CIA Kryptos Puzzle Creator Releases Final Clues”, Scientific American, November 2025; and “This Puzzle Has Been Unsolved For 35 Years—Now We Have the Final Clues”, Popular Mechanics, November 2025, for K5.
- Keith Adler, keithadler/kryptos.
- Moshe Rubin, “Chaocipher Revealed: The Algorithm”, 2010.