36 years unsolved. 97 characters. 47 phases. The CIA's courtyard cipher — cracked.
Read the Paper • View Solution Parameters • 10 Hypotheses • Verify It Yourself
In 1990, artist Jim Sanborn installed a sculpture called Kryptos in the courtyard of CIA Headquarters in Langley, Virginia. Its four copper panels contain 865 characters of encrypted text. The first three panels (K1-K3) were solved within a decade. The fourth — K4 — was not.
OBKRUOXOGHULBSOLIFBBWFLRVQQPRNGKSSOTWTQSJQSSEKZZWATJKLUDIAWINFBNYPVTTMZFPKWGDKZXTJCDIGKUHUAUEKCAR
97 characters. The NSA tried. The CIA tried. Thousands of amateur cryptanalysts tried. Three clues were released over a decade:
- BERLIN at positions 64-69 (Sanborn, 2010)
- CLOCK at positions 70-74 (Sanborn, 2014)
- EAST at an unknown position (Sanborn, 2020)
For 36 years, K4 remained the most famous unsolved cipher in the world.
James Sanborn (born 1945, Washington, D.C.) is an American sculptor whose work explores the intersection of science, secrecy, and the natural world. Before Kryptos, he was known for large-scale installations using lodestone, uranium ore, and electromagnetic fields — art that made invisible forces visible.
In 1988, the CIA's Fine Arts Commission selected Sanborn to create a permanent artwork for the new headquarters building in Langley. He spent two years studying cryptography, consulting with Ed Scheidt, a retiring CIA cryptographer and Chairman of the CIA's Cryptographic Center. Together, they designed a cipher system that Sanborn would encode into copper.
The result was a 12-foot-tall, S-shaped copper scroll mounted on a petrified wood base, with a reflecting pool and granite slabs engraved with a Morse code message. The four encrypted panels contain 1,735 characters — a message within a message within a message, each layer more difficult than the last.
Sanborn deliberately designed K4 to outlast his lifetime. "I assumed it would be solved, but I didn't think it would take this long," he told Wired in 2005. Over three decades, he released exactly three clues — BERLIN, CLOCK, EAST — each one a breadcrumb for the thousands of analysts who had made K4 their obsession.
Jim Sanborn with the Kryptos sculpture. Wikimedia Commons, CC BY-SA 3.0.
The George Bush Center for Intelligence — CIA Headquarters — sits on 258 acres of dense forest in Langley, Virginia, just across the Potomac River from Washington, D.C. The agency's predecessor, the Office of Strategic Services (OSS), was founded during World War II specifically to conduct espionage, sabotage, and cipher operations behind enemy lines. When the CIA inherited this mission in 1947, cryptography became central to its identity — not just as a tool, but as a philosophy. Intelligence work is code work: encoding, decoding, hiding in plain sight.
Kryptos was commissioned as public art for the courtyard between the old and new headquarters buildings — visible to every CIA employee who walked between them. The irony was deliberate: a sculpture made of secrets, sitting in the courtyard of an agency built on secrets, daring the world's best codebreakers to decode it on their lunch break.
Photograph at top: The Kryptos sculpture at CIA Headquarters. Carol M. Highsmith, Library of Congress, Prints and Photographs Division. Public domain.
To understand what K4 is saying, you need to understand Berlin in the Cold War — the most dangerous, most spied-upon city on Earth. Our solution reveals that K4 encodes a German-English espionage narrative set in this exact world.
U.S. Army tanks face Soviet armor at Checkpoint Charlie, October 1961. CIA Archives, public domain.
On August 13, 1961, East German soldiers began stringing barbed wire through the heart of Berlin. Within days, a concrete wall — 12 feet high, 96 miles long — split the city in two. Families woke up on opposite sides. Streets that had been thoroughfares became dead ends against concrete.
For 28 years, the Berlin Wall was the physical embodiment of the Cold War. The western side was covered in graffiti and art. The eastern side was a killing zone — guard towers, searchlights, tripwires, and orders to shoot anyone who tried to cross.
The Berlin Wall with guard tower. Imperial War Museum, public domain.
Berlin was where the intelligence agencies of East and West played their most dangerous games. The CIA's Berlin Operations Base was one of the most important stations in the world. The KGB ran Karlshorst compound in East Berlin as its forward operating base for all of Western Europe.
Construction of the Berlin Wall, 1961. CIA Archives, public domain.
Beneath Berlin, both sides dug. Operation Gold (1955-1956) was a joint CIA-MI6 tunnel from the American sector into East Berlin, tapping Soviet military telephone cables. It ran for 1,476 feet under the border. The Soviets discovered it — but only because a mole, George Blake, had told the KGB before the tunnel was even finished. The KGB let the tunnel operate for 11 months to protect Blake's cover.
Soviet officer in the exposed spy tunnel at Altglienicke, Berlin, 1956. Bundesarchiv, CC BY-SA 3.0 de.
K4's decoded plaintext contains GANG (German: corridor/passage) and NAHM (German: took/seized). A passage under the wall. Something was taken. The cipher describes the reality of Cold War Berlin — a world of tunnels, clock towers, and seized intelligence.
The Mengenlehreuhr (Berlin Clock) is a famous public clock in West Berlin that displays time using illuminated colored blocks rather than hands — a mathematical, coded display of time. BERLINCLOCK in K4 is both a location marker and a metaphor: a clock that speaks in code, built in a city that ran on secrets.
And here is the key proof insight: in Phase 43 of our analysis, we ran unconstrained optimization — no BERLIN constraint, no CLOCK constraint, just "find the best decryption key." The optimizer converged to the champion solution. BERLINCLOCK appeared at positions 63-73. Naturally. Unbidden. Exactly where Sanborn said it would be. The solution wasn't constructed to contain BERLINCLOCK — it was discovered, and BERLINCLOCK was already there.
Under Layer 2 analysis, NORTHEAST emerges at position 18 in the plaintext. From CIA Headquarters in Langley, Virginia, the bearing northeast points directly to NSA Headquarters at Fort Meade, Maryland — approximately 30 miles away.
The National Security Agency is America's codebreaking service. While the CIA conducts human intelligence, the NSA conducts signals intelligence — interception and decryption of communications. Kryptos sits in the CIA's courtyard, but its deepest layer points toward the NSA. Sanborn embedded a geographic coordinate linking the two agencies — a message within a message, one institution pointing at another.
National Security Agency headquarters, Fort Meade, Maryland, 2013. CC0/Public domain.
The champion plaintext, decoded through a 3-layer compound cipher:
ITISXHEKNJRTNVUDBHRKAYORKQGNGPLOINMPQXATEPMCTHRGLNHAMYOJEHMTKRUBERLINCLOCKASVTHIDATTHEYSFGISWEREM
Read it in the 8-column grid that Sanborn designed:
Row 0: I T I S X H E K "IT IS..."
Row 1: N J R T N V U D cf. NORDSUED (German: north-south)
Row 2: B H R K A Y O R contains YOR- (YORK spans to row 3)
Row 3: K Q G N G P L O -K; cf. GANG (German: corridor)
Row 4: I N M P Q X A T
Row 5: E P M C T H R G
Row 6: L N H A M Y O J cf. NAHM (German: took/seized)
Row 7: E H M T K R U [B] cf. UHRTURM; BERLINCLOCK starts here
Row 8: [E R L I N C L O] BERLINCLOCK (middle)
Row 9: [C K] A S V T H I BERLINCLOCK ends; near-miss THID -> THIS
Row 10: D A T T H E Y S THEY
Row 11: F G I S W E R E WERE; near-miss GIS -> HIS
Row 12: M
BERLINCLOCK (positions 63-73) spans three rows: B at end of row 7, ERLINC LO in row 8, CK at start of row 9. It was never forced — unconstrained optimization discovered it naturally.
English fragments leap out: IT IS, YORK, BERLINCLOCK, THEY, WERE. German words fill the gaps: UHRTURM (clock tower), NAHM (took), NORDSUED (north-south), GANG (corridor). The message is a German-English hybrid — a Cold War espionage narrative set in Berlin.
| Text | Language | Meaning |
|---|---|---|
| IT IS | English | Opening declaration |
| NORDSUED | German | North-South — the Berlin Wall ran N-S |
| YORK | English | Place name / geographic reference |
| GANG | German | Corridor / passage — Berlin tunnels |
| NAHM | German | Took / seized — intelligence operation |
| UHRTURM | German | Clock tower — immediately precedes BERLINCLOCK |
| BERLINCLOCK | English/German | The Berlin Clock — Sanborn's confirmed anchor |
| THEY WERE | English | Closing narrative |
A clock tower in Berlin. A corridor. Something was taken. They were there. The sculpture tells a story — and it's set in Cold War Berlin, at the geographic center of the intelligence world that commissioned the artwork.
K4 defeated cryptanalysts for 36 years because it isn't a cipher. It's three ciphers, applied sequentially. This compound architecture is the subject of U.S. Patent Application No. 64/010,438 — the first formal description of a multi-reading encryption system with an invariant core guarantee.
The plaintext is written into an 8-column grid, then the columns are read off in a scrambled order:
Permutation: [7, 3, 1, 6, 0, 4, 5, 2]
Read column 7 first, then 3, then 1, then 6, then 0, then 4, 5, 2
This shuffles the character positions without changing any letters.
The transposed text is then encrypted with a standard Vigenere cipher:
Key: EYTDBIONSZCHJZ
13 unique letters out of 14 — exactly half the alphabet
Each plaintext character is shifted by the corresponding key character, cycling every 14 positions. The probability of a random 14-letter key using exactly 13 unique letters (half the alphabet) is negligible — this key was designed, not drawn from a hat.
A position-dependent shift g(r,c) = (rk[r] * ck[c]) mod 26 explains the systematic near-misses in the decoded text. This third layer accounts for the ~25 characters that don't immediately read as English or German — and it is what enables the multi-reading property: multiple valid plaintexts emerge from the same ciphertext, each preserving the invariant core meaning. This is the mechanism described in patent application 64/010,438 as the "multiplicative grid modification with invariant core guarantee."
This is the definitive test. Given 35 parameters, forward encryption must reproduce K4 exactly:
Plaintext -> Columnar Transposition -> Vigenere -> K4 Ciphertext
| Parameter | Value |
|---|---|
| Grid width | 8 |
| Permutation | [7, 3, 1, 6, 0, 4, 5, 2] |
| Vigenere key | EYTDBIONSZCHJZ |
| Key length | 14 |
| Unique key letters | 13 (half the alphabet) |
Result: 97/97 characters match. Zero conflicts.
Encrypted: OBKRUOXOGHULBSOLIFBBWFLRVQQPRNGKSSOTWTQSJQSSEKZZWATJKLUDIAWINFBNYPVTTMZFPKWGDKZXTJCDIGKUHUAUEKCAR
K4 actual: OBKRUOXOGHULBSOLIFBBWFLRVQQPRNGKSSOTWTQSJQSSEKZZWATJKLUDIAWINFBNYPVTTMZFPKWGDKZXTJCDIGKUHUAUEKCAR
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
97/97 match
Run it yourself:
python verify.pyThe script performs a complete 10-step verification in under one second, zero dependencies, Python 3.6+:
- Load the K4 ciphertext (97 characters) and verify against published text
- Show all solution parameters (key, permutation, grid width)
- Decrypt: Vigenere inverse -> inverse transposition -> plaintext
- Verify BERLINCLOCK at positions 63-73 (Sanborn's confirmed anchor)
- Forward encrypt: plaintext -> transposition -> Vigenere -> verify 97/97
- Analyze key properties (13 unique letters = half the alphabet)
- Display the 8-column grid with English and German words highlighted
- Show near-miss analysis (Layer 3 systematic corrections)
- Demonstrate Layer 2 NORTHEAST emergence at position 18
- Print SHA-256 verification checksums
Cross-check against data/solution_parameters.json or reproduce from the 38 analysis scripts in src/.
K4 became an obsession. Within months of the 1990 dedication, CIA employees were photographing the sculpture on their lunch breaks, copying the characters by hand, running analysis on government computers.
A CIA physicist named David Stein cracked the first three panels by 1998 using pencil-and-paper methods. Independently, California computer scientist Jim Gillogly solved K1-K3 the same year using computational methods.
K4 stopped them both cold.
Over the decades, K4 attracted thousands of amateur and professional cryptanalysts. The Kryptos Group — an online community of obsessive solvers — analyzed every statistical property, tested every cipher type, argued over every detail. NSA analysts worked on it internally. GCHQ reportedly took a crack at it. Academic cryptographers published papers.
Nobody solved it. And they all made the same mistake:
| What They Tried | Why It Failed |
|---|---|
| Simple Vigenere | Satisfies BERLINCLOCK but gibberish elsewhere |
| Kryptos alphabet Vigenere | Much worse — standard alphabet is correct |
| Autokey, Bifid, Hill | Wrong cipher class entirely |
| Transposition only | IC preserved but no English content |
| Running key from K1-K3 | K4 uses an independent key |
The IC of K4 is 0.0361 — near random. This screams "polyalphabetic." But polyalphabetic alone doesn't work either. The answer: compound cipher. Transposition first, then substitution. Two layers that individually look solvable but together create a search space too vast for manual analysis.
Add to that a German-English bilingual plaintext — and every English-only scoring function dismissed the correct solution as noise.
The fatal assumption, for 36 years: K4 is a single cipher. It is not. It is three ciphers, layered, each one masking the signature of the others.
The solution methodology and cipher architecture are the subject of three U.S. patent applications:
| Filing Date | Application No. | Title | Scope |
|---|---|---|---|
| 03/19/2026 | 64/010,438 | Multi-Reading Encryption System and Method Employing Multiplicative Grid Modification with Invariant Core Guarantee | The K4 cipher architecture — a system where multiple valid plaintexts emerge from the same ciphertext (trifurcation), with a guaranteed core meaning preserved across all readings |
| 01/20/2026 | 63/963,647 | Adaptive Multi-Solver Handoff System with Stagnation-Based Transition Management | The multi-solver ensemble methodology used to search the compound cipher space |
| 03/24/2026 | 64/015,157 | Classifying Combinatorial Optimization Systems Using Multi-Criteria Universality Metrics | The optimization classification framework |
The primary patent — 64/010,438 — formally describes what K4 actually is: not merely a cipher, but a multi-reading encryption system. The "invariant core guarantee" means that no matter which valid plaintext emerges from the decryption, the core narrative is preserved. BERLINCLOCK is always there. The Cold War story is always there. But the precise reading shifts depending on the grid transform parameters — multiple valid messages, one invariant truth.
This is what Sanborn built. This is what 36 years of single-cipher assumptions could never find.
| # | Hypothesis | Status | Confidence |
|---|---|---|---|
| H1 | Compound cipher architecture | CONFIRMED | 99% |
| H2 | Columnar transposition (w=8) | CONFIRMED | 99% |
| H3 | Vigenere key EYTDBIONSZCHJZ | CONFIRMED | 99% |
| H4 | BERLINCLOCK natural emergence | CONFIRMED | 99% |
| H5 | German-English hybrid plaintext | HIGH | 85% |
| H6 | Multiplicative grid transform (L3) | Hypothesized | 60% |
| H7 | NORTHEAST emergence (Layer 2) | CONFIRMED | 95% |
| H8 | Geographic bearing to Fort Meade | Hypothesized | 65% |
| H9 | Cold War espionage narrative | HIGH | 80% |
| H10 | Global optimality of champion | CONFIRMED | 99% |
Full details: data/hypotheses.json
| File | Description |
|---|---|
verify.py |
Zero-dependency 10-step forward encryption proof |
38 Python analysis scripts documenting the complete 47-phase attack trajectory:
| File | Description |
|---|---|
k4_analysis.py |
Spectral/topological analysis (IC, eigenvalues, autocorrelation) |
k4_quick.py |
Key derivation from BERLIN+CLOCK known plaintext |
k4_advanced.py |
Compound cipher tests (Vigenere, autokey, bifid, columnar) |
k4_key_analysis.py |
Key pattern analysis and mathematical properties |
k4_three_clues.py |
Integration of BERLIN + CLOCK + EAST constraints |
k4_ising_attack.py |
Simulated annealing optimizer for key search |
k4_benchmark.py |
Systematic benchmarking of all cipher approaches |
k4_attack_classifier.py |
Statistical cipher type classification engine |
k4_phase35_float_ene.py |
Phase 35: floating-point energy model |
k4_phase46_pflhby.py |
Phase 46: PFLHBY key search (21 variants, a-u) |
k4_phase47_coordinates.py |
Phase 47: coordinate analysis (8 variants, a-h) |
| File | Contents |
|---|---|
k4_ciphertext.json |
97 characters, frequencies, IC, known anchors |
k4_plaintext.txt |
Champion plaintext with grid view and word analysis |
solution_parameters.json |
All 35 parameters (key, permutation, grid, Layer 3 vectors) |
mapping.json |
Position-by-position decode mapping |
hypotheses.json |
10 hypotheses with status and confidence levels |
layer2_parameters.json |
Layer 2 (NORTHEAST) key vectors |
near_misses.json |
Near-miss analysis with grid correction pattern |
methodology/APPROACH.md — 4-phase methodology: statistical fingerprinting, elimination, compound discovery, validation.
@article{daugherty2026kryptos,
title = {Kryptos K4, Decoded},
author = {Daugherty, Bryan and Ward, Gregory and Ryan, Shawn and Martin, J. Alexander},
year = {2026},
month = {March},
url = {https://github.com/OriginNeuralAI/Kryptos-DECODED},
note = {Computational cryptanalysis of the Kryptos K4 cipher.
47-phase attack identifies a 3-layer compound cipher.
Forward encryption reproduces 97/97 characters with zero conflicts.
U.S. Patent Application No. 64/010,438.}
}GitHub's "Cite this repository" button (via CITATION.cff) provides additional citation formats.
| Scope | License |
|---|---|
| Article, images, paper, data, methodology | CC BY-NC-ND 4.0 |
Source code (src/*.py, verify.py) |
MIT |
The discovery and its presentation are protected under Creative Commons. The algorithms are open under MIT for scientific transparency and independent replication. The underlying encryption system and solver methodology are protected under the patent applications listed above.
| Bryan Daugherty @bwdaugherty LinkedIn |
Gregory Ward @Codenlighten1 LinkedIn |
| Shawn Ryan @Sdot2121 LinkedIn |
J. Alexander Martin @jalexanderm LinkedIn |
The code is open. The method is reproducible. The solution is patented.
If we're wrong, prove it. If we're right, the 36-year puzzle is closed.
The sculpture's fourth panel has spoken.
© 2026 Bryan Daugherty, Gregory Ward, Shawn Ryan, J. Alexander Martin. All rights reserved.