PURPLE — Volume 2 — Before PURPLE — Japan's Codes, Yardley's Black Chamber and the Naval Treaty

2.1 About This Volume

PURPLE was not the first Japanese cipher the Americans read, and the machine exists partly because they had read the earlier ones. This volume covers the twenty years before the Type 97 machine went into service in February 1939: the codebooks Japan’s diplomats used, the small and secret American bureau that broke them, the 1921–22 naval conference at which that bureau’s work paid off spectacularly, the bureau’s abrupt closure in 1929, and the 1931 book in which its former chief told the world — and Japan — what it had done.

It also follows the parallel story in the U.S. Navy, whose agents photographed a Japanese naval codebook in New York, and the founding in 1930 of the U.S. Army’s Signal Intelligence Service, the team that would later break both RED (Volume 3) and PURPLE (Volume 5). The volume ends where Japan’s answer begins: the decision to move its most important traffic from codebooks to machines.

2.2 Japan on the Wire

A Japanese diplomat in Washington, London or Berlin in the 1920s communicated with Tokyo by commercial cable. The cable companies carried messages in the Latin alphabet, so Japanese text — normally written in a mixture of Chinese characters and the kana syllabaries — was first spelled out in Latin letters, in one of the romanisation systems known as rōmaji. A message for secrecy was then put into code, a book of numbered or lettered groups standing for syllables, words and set phrases, and often enciphered on top of that with tables that changed periodically. What went down the wire was a stream of letter groups, usually five letters long, which any clerk could transmit and which no one outside the Foreign Ministry could read without the books.

That chain explains several features of the story to come. Codebooks are expensive to replace, so a broken book could be read for months or years. Romanised Japanese has a distinctive statistical texture — a regular alternation of consonants and vowels, frequent syllables such as no, ni and wo, and recurring set phrases — which a cryptanalyst can learn to recognise even inside a cipher. And because the cable companies and the governments that licensed them could see every group that passed, any state with access to the cable offices could collect a Japanese embassy’s traffic in full.

Japan took cryptography seriously and was itself learning to break codes. Polish sources record that Captain Jan Kowalewski, the officer who had led Poland’s radio-intelligence work against the Red Army in 1919–20, ran a cryptology course for Japanese Army officers in Tokyo in the first half of 1923 and was decorated for it (Polish History, “Jan Kowalewski’s three victories”). But the Foreign Ministry’s own systems in the early 1920s were codebooks of a conventional kind, and the first service to read them systematically was American.

2.3 The Cipher Bureau

Herbert Osborne Yardley was born in Worthington, Indiana, in 1889, learned telegraphy from his father, a railroad station master, and became a code clerk at the State Department in 1912. He taught himself to solve the department’s own codes and in May 1916 wrote a long memorandum, “Solution of American Diplomatic Codes”, showing how weak they were. When the United States entered the First World War he was commissioned in the Signal Corps and in 1917 put at the head of a new military-intelligence section for codes and ciphers, MI-8 (NSA, The Many Lives of Herbert O. Yardley).

After the war the Army and the State Department agreed to keep a codebreaking unit alive in peacetime, funding it jointly and secretly. To keep it away from Washington’s scrutiny and from legal questions about reading foreign governments’ cables, it was set up in New York in 1919, first in a brownstone at 3 East 38th Street and from 1 July 1920 at 141 East 37th Street, with Yardley and his wife living on one floor. Staff were paid in cash from a secret payroll and told to describe themselves as War Department translators. As a front, Yardley and the linguist Charles Mendelsohn ran a business called the Code Compiling Company, which sold commercial code books. The State Department’s monthly contribution ranged from $1,250 to $3,333 and Military Intelligence’s was a steady $833.33, giving the bureau between $15,000 and $50,000 a year. Its traffic came from the commercial telegraph companies — Western Union and Postal Telegraph among them — under informal arrangements with the government.

The bureau was officially the Cipher Bureau; its staff and its chief called it the Black Chamber, after the cabinets noirs where European post offices had once opened diplomatic mail.

2.3.1 A promise and a Japanese code

In July 1919 Yardley promised his superior, Brigadier General Marlborough Churchill, the Director of Military Intelligence, that he would solve the Japanese diplomatic code within a year or resign. Neither he nor anyone on his staff knew Japanese. Within five months he had identified enough code groups to know he had the right solution, and wrote to Churchill:

I have passed through various periods of confidence and depression, but it was not until Saturday morning, about One A.M., that I locked my safe, and with it the correct solution of the code… With the aid of a good Japanese scholar there is no doubt but that I can have the Japanese code complete for you and probably some important messages before you go to Congress for the MID appropriation… With the exception of clerical assistance I have worked practically alone, and it is the first thing that I have ever done which I feel really proud of.

— Yardley to Churchill, quoted in The Many Lives of Herbert O. Yardley

The bureau hired a former missionary, the Reverend I. H. Correll, as its Japanese linguist, and with his help translated its first Japanese message in February 1920. Military Intelligence called it the “most remarkable accomplishment in the history of code and cipher work in the United States.” When Correll left, one of the analysts, Frederick Livesey, who had learned Japanese in six months on top of his other languages, took over. Japanese communications became the bureau’s first priority — diplomatic codes and, increasingly, the military and naval attaché codes as well.

2.4 The Washington Naval Conference

In the summer of 1921 the United States invited the major naval powers to Washington to negotiate limits on warships. The Conference on the Limitation of Armament opened on 12 November 1921 at Memorial Continental Hall, and on its first day the American Secretary of State, Charles Evans Hughes, astonished the delegates by proposing that the powers scrap dozens of capital ships, built and building, and accept fixed ratios of battleship tonnage. The question that mattered most to the Pacific was Japan’s ratio. Hughes proposed 5:5:3 for the United States, Britain and Japan. The Japanese delegation, led by the Navy Minister, Admiral Baron Katō Tomosaburō, with the ambassador Shidehara Kijūrō and Prince Tokugawa Iesato, insisted on 10:7, or 70 per cent of the American figure.

Through the conference the Black Chamber read the telegrams between Tokyo and the delegation. Yardley called one of them “the most important and far-reaching telegram” that ever passed through his bureau. It was message number 700, Tokyo to Washington, dated 28 November 1921 and addressed to Admiral Katō; the bureau received, decoded and translated it on 2 December. Wishing to “avoid any clash with Great Britain and America, especially America, in regard to the armament limitation question,” Tokyo authorised its delegates, if pressed, to accept the American ratio. Hughes had only to hold his position. Japan accepted 5:5:3 about a week later. Yardley, a serious poker player, summed it up: “Stud poker is not a very difficult game after you see your opponent’s hole card.”

The Five-Power Treaty, signed on 6 February 1922, fixed capital-ship tonnage at 525,000 tons each for the United States and the British Empire and 315,000 tons for Japan, with lower figures for France and Italy. The NSA’s own history is careful about the decrypts’ value: some were timely and revealed the Japanese government’s internal debates and dissents over its position; others arrived late and repeated what the newspapers had already printed. The cost to the bureau was considerable. Its fourteen staff worked nights and weekends for the three months of the conference and produced several thousand translations; at Christmas 1921 they shared a bonus of between $37 and $184 each, paid out of the bureau’s unaudited funds, and Yardley left for Arizona to recover from a mild case of tuberculosis. Secretary Hughes wrote him a letter of appreciation, and early in 1923 he received the Distinguished Service Medal, with a citation that said nothing about Japan.

2.5 “Gentlemen Do Not Read Each Other’s Mail”

The conference was the Black Chamber’s high point. A budget cut in 1924 reduced it by more than half; the six who remained — Yardley, two analysts, two clerks and a secretary — moved to an office at 52 Vanderbilt Avenue. The telegraph companies grew reluctant to hand over copies of cables, and the Radio Act of 1927 may have strengthened their reluctance. Yardley later hinted at “rather subtle methods” of obtaining traffic; no record supports him. The bureau still reported successes against Japanese, Mexican, Nicaraguan, Peruvian and Chinese systems, and in its twelve years it read the messages of some thirty-five to forty countries.

The end came in 1929. Henry L. Stimson became Secretary of State on 28 March. According to what the NSA history calls the “Standard Version”, when he learned what Yardley’s bureau was doing he withdrew the State Department’s funding on the grounds that the activity was unethical. Since State supplied most of the money, the bureau could not survive; it ceased work in the early summer and closed officially at the end of October, two days after the stock-market crash. Stimson himself, in the memoir he wrote with McGeorge Bundy in 1948, On Active Service in Peace and War, gave the sentence that has followed the episode ever since: “Gentlemen do not read each other’s mail.”

The story is neater than the record. The Army had begun in March 1929 — before Stimson took office — to move responsibility for codebreaking from Military Intelligence to the Signal Corps, which already compiled the Army’s codes, and the change was approved by the Secretary of War on 4 April and written into Army Regulations on 10 May. Yardley came to believe that Stimson had acted on President Hoover’s orders; Thomas Dewey later repeated the claim. Whatever the motives, the bureau’s staff were offered Signal Corps posts in Washington at lower salaries and declined them, and Yardley, with no civil-service status and no pension, went home to Indiana as the Depression began (Many Lives). The irony is that Stimson, who closed America’s diplomatic codebreaking in 1929, would be Secretary of War in 1940 and one of the first men shown the machine that read PURPLE (Volume 6).

2.6 The Navy’s Back Door

The Army’s bureau was not the only American effort against Japan. The U.S. Navy cared less about diplomatic codes than about the Imperial Japanese Navy’s own systems, and at least once it took a shortcut that its historians call “practical cryptanalysis”. Agents of the Office of Naval Intelligence entered Japanese offices in New York, opened a safe and photographed a Japanese naval codebook. Accounts of the date differ: veterans’ histories place the first entry around 1920, while one researcher puts it between 1920 and 1923 and identifies a better-documented entry in 1929. The result was what the Navy called the RED BOOK — named, the NSA history notes, for the colour of the binder, and not to be confused with the RED machine of Volume 3. It contained some 100,000 entries, was translated by a missionary couple, and in 1929 was retyped in four copies, one of them for the Asiatic Fleet (NSA, “Early Japanese Systems”).

A codebook is not enough by itself: the messages were also enciphered. Lieutenant Joseph Rochefort and Agnes Meyer Driscoll, a mathematician who had joined the Navy as a yeoman in 1918 and become its outstanding cryptanalyst, solved enough RED BOOK traffic to reveal Japanese naval training and supply. In December 1930 Japan replaced the book with the BLUE BOOK, about 85,000 groups, and this time there was no burglary: Lieutenant Thomas Dyer broke it the hard way, after a key suggestion from Driscoll. BLUE BOOK messages showed that Japan’s newest battleships could make more than 26 knots, two knots faster than the American battleships then building, and the U.S. Navy designed its next generation for 27. The Navy’s small codebreaking section, designated OP-20-G and headed for long stretches by Laurance Safford, grew out of this work. Its rivalry and cooperation with the Army’s section shaped everything that followed, including the sharing of PURPLE.

2.7 Friedman Starts Again

While Yardley’s bureau was ending, the Signal Corps was planning its replacement. Its chief cryptologist was William F. Friedman, born in Russia in 1891, trained as a geneticist at Cornell, drawn into cryptology at the Riverbank Laboratories in Illinois and sent to France in 1918 to work on German codes. Friedman had been the Signal Corps’ civilian code compiler since 1922. In 1929 the Army decided to create a Signal Intelligence Service (SIS) with Friedman as its chief civilian, charged with compiling the Army’s codes and training cryptanalysts for war — not, officially, with reading anyone’s traffic in peacetime (NSA, “Signal Intelligence Service”).

On 1 April 1930 — All Fools’ Day, as he liked to point out — a mathematics teacher, not quite 22, from southern Virginia named Frank B. Rowlett reported to Room 3406 of the Munitions Building in Washington as the first of three “junior cryptanalysts” at $2,000 a year. Two former New York mathematics teachers, Abraham Sinkov and Solomon Kullback, followed later that month, and soon afterwards a Japanese linguist, John B. Hurt, who had learned the language rooming with Japanese students at college. Friedman trained them on problems of his own devising and then, one hot morning in June 1930, led them down a deserted corridor, unlocked a vault, lit a match and introduced them to its contents. The National Cryptologic Museum’s account gives his words as “Welcome, gentlemen, to the archives of the American Black Chamber” (NSA, “The Magic of Purple”). The files of Yardley’s bureau had come to Washington in October 1929, and Friedman’s new hires spent weeks learning from them how the Japanese and other systems had been solved.

The new section was tiny — seven people at its centre from 1930 to 1936, nineteen by 1939 — and at first it worked on Japanese traffic only as training. But it was building, deliberately, a capacity against cipher machines. Friedman had his staff study and solve the rudimentary commercial machines of the day, because all of them understood, as the NSA history puts it, that machines were the future of cryptography. They were right sooner than they expected.

2.8 A Bombshell in 375 Pages

Yardley needed money. He wrote a book. The American Black Chamber appeared on 1 June 1931, preceded by excerpts in the Saturday Evening Post, and it told nearly everything: the organisation, its methods, its budget and its successes, illustrated with real intercepts and translations — of allies as well as adversaries. It was dedicated “to Our Skillful Antagonists, The Foreign Cryptographers, who still remain behind the Curtain of Secret Diplomacy.” The book sold 17,931 copies in the United States and 5,480 in Britain and was translated into French, Swedish, Japanese and Chinese.

The reaction in Washington was fury mixed with helplessness. The State Department disavowed the bureau; Stimson’s officials denied that Yardley had broken Japanese codes during the conference at all. Friedman led a damage assessment that found the book a breach of confidence “full of distortions and exaggerations.” No law clearly forbade what Yardley had done. In 1933, when he tried to publish a second book, Japanese Diplomatic Secrets, the manuscript was seized, and Congress passed the law, now 18 U.S.C. § 952, that makes it a crime for a government employee to publish foreign diplomatic codes or messages obtained in transmission.

The reaction in Japan was greater. The Japanese translation sold, according to one count, 33,119 copies, and the Osaka Mainichi and Tokyo Nichi-Nichi serialised it with interviews with the author. Headlines spoke of “Betrayal of International Trust” and “Treachery at the Washington Conference”. The government’s opponents blamed Shidehara — by 1931 Foreign Minister — for failing to change the codes, and the controversy fed a wave of anti-American feeling in the year of the Manchurian Incident, when relations were already strained. In August 1931 the Foreign Ministry’s communications department asked permission to buy 138 copies of the book for distribution to Japan’s embassies and legations, so that every post would know how exposed its codes had been (Many Lives).

Yardley’s reputation never recovered. Ladislas Farago claimed in 1967 that Yardley had sold the Japanese embassy his bureau’s secrets for $7,000 before he wrote the book; a 1967 CIA memorandum dated the sale to June 1930. The Japanese document Farago said he had seen has not been found, and historians remain divided (Knox College). What is beyond dispute is the book’s effect.

2.9 From Codebooks to Machines

Japan had begun experimenting with cipher machines before Yardley published, but The American Black Chamber gave the work urgency. The NSA’s historians, drawing on the Navy cryptanalyst Francis Raven’s post-war notes, record that in the early 1930s the Japanese Navy approached the Swedish cipher-machine maker Boris Hagelin, who — wary that his designs would be copied — sold only some obsolete machines invented by his late partner Arvid Damm. The Japanese studied them and built their own family of devices, designated Type 91 (91-shiki) after the Japanese imperial year 2591, which is 1931. The Navy’s version became the naval attaché machine the Americans called M-1 or “Naval Attaché RED”, which Driscoll solved in 1934 or 1935. The Foreign Ministry’s version became the diplomatic machine the SIS called RED.

The Many Lives history puts the sequence plainly: aware, thanks to Yardley, of how insecure their communications had been in 1921, the Japanese were determined to use cipher machines at the next disarmament conference, due to open at Geneva early in 1932. In October 1931 the U.S. Navy reported that it could no longer obtain “information regarding the naval activity of a foreign power” as it had, because “the systems of conveying this information have undergone a complete change,” and that reconstructing them would require “the entire efforts of our personnel for a period of at least two years.”

The first diplomatic machine carried a fingerprint of the telegraph age. International cable tariffs at the time charged less for “pronounceable” text — groups containing at least two vowels in five letters — than for unpronounceable code. RED therefore enciphered vowels as vowels and consonants as consonants, so that its output stayed cheap to send. When SIS began working on RED intercepts it noticed exactly that: two classes of letters, six and twenty, with different frequencies. The tariff concession ended in March 1934 and the machine’s division was rearranged, but the six-and-twenty split survived into PURPLE, where it became the crack through which SIS entered (Volume 4).

By the mid-1930s, then, the board was set. Japan had replaced its most important codebooks with machines it believed unreadable. The United States had a small, well-trained Army section that had studied machine ciphers for five years, a Navy section with its own Japanese experience, and an uneasy habit of cooperation between the two. Across the Pacific and the Atlantic, traffic passing through American cable offices and radio intercept stations was quietly collected. Volume 3 picks up the story in 1936, when Rowlett and Kullback turned to RED.

---

Next — Volume 3: RED — The Type 91 Machine and Its Break. Japan’s first diplomatic cipher machine: a half-rotor that kept vowels among vowels, a break wheel of 47 positions that stepped it irregularly, and contacts that had to be cleaned every day. How Frank Rowlett and Solomon Kullback broke it from intercepts alone in 1936, why they named it after the first colour of the spectrum — and how RED itself warned them, in late 1938, that a Type B machine was on its way.