Code Warriors

by Stephen Budiansky

  Attempts to effect a postwar merger of the Army and Navy signals intelligence units had meanwhile dissolved into an acrimonious bureaucratic struggle that continued through the fall and winter of 1945. The Army strongly favored the idea, and some in the Navy did, too, arguing that a combined organization would be better able to resist outside control, avoid wasteful duplication, bring strength through numbers to bear on the most important problems, and prevent a repetition of the often absurd situation that had arisen during the war when the British, with whom each of the American operations often had better contacts than they did with each other, ended up acting as the go-between, allowing GC&CS to “play one service against the other” for its own purposes, Wenger noted.59

  But Admiral King hewed firmly to the traditional line that the Navy had to be in full command of its own operations and not allow a situation ever to arise in which some civilian or general could tell them what to do. Op-20-G was in any case insisting that its 1942 agreement to hand over all work on diplomatic codes to the Army had been merely a “wartime expedient” (and was “invalid” in any case), and that the work should once again be shared now that the volume of military and naval traffic was rapidly dwindling. The Army was perfectly amenable to that, but on the larger issue of postwar organization and control King and Marshall were at a complete impasse. Wenger confided to a colleague in February 1946 that “everything has been in a state of confusion here” owing to the unresolved Army-Navy differences and “not a single basic decision affecting our future” had been made while the battle raged.60

  The deeper anxiety that gnawed at both the British and American codebreakers was how fragile the whole structure was. Without a steady stream of coded material to work on, it would be extremely difficult to maintain either the prestige or the proficiency of the organizations, even if their personnel and budget problems were solved. Successful breaking of a code system required above all what the cryptanalysts called “continuity”: if work was dropped even for a short time it often required a huge research effort to recover lost ground. And William F. Clarke, a veteran of the GC&CS naval section whose experience stretched back to the First World War and who still recalled with undiminished bitterness the blockheaded attitudes he had frequently encountered from naval officers over the years and the endless battles that GC&CS had had to fight to “convince the doubters of its necessity,” gloomily foresaw more of the same in the coming years, especially when “those at the moment in the highest places” of government who had come to fully understand the value of signals intelligence were replaced by others without wartime experience in its use.61

  But February 1946 suddenly brought a flurry of developments that profoundly shaped the course of the postwar signals intelligence establishment. The Army and Navy agreed to a cease-fire in their dispute, with an agreement to keep their two establishments separate for now but with a rotating “coordinator” who would allocate tasks that fell within their joint responsibility for diplomatic codes, and with a new interagency board that included a representative of the State Department to set communications intelligence policy. After months of bureaucratic wrangling over the overall government intelligence structure, President Truman named the first director of central intelligence to coordinate all U.S. intelligence activities and head an independent board that, the following year, would become the CIA—a full-fledged agency in its own right. That same month the new State-Army-Navy Communications Intelligence Board and GC&CS reached a formal, and breathtakingly comprehensive, British-U.S. Communications Intelligence Agreement that provided for the “unrestricted exchange” of “all work” undertaken by each, considerably expanding their close wartime collaboration into the uncertain peace ahead. And in a rare public speech delivered with much fanfare at Moscow’s Bolshoi Theater on February 9, Joseph Stalin declared that conflict between the Communist and capitalist states was inevitable, that the Soviet Union must triple its prewar levels of industrial production to “guarantee our country against any eventuality,” and that the people must be prepared to make sacrifices as production of consumer goods “must wait on rearmament.”

  Stalin actually was saying nothing he had not said before, and his long didactic explanation of capitalism’s tendency to produce war was straight out of classic Marxist-Leninist theory, but his words stunned Washington. For many it was the first realization that Stalin’s vision of peace had literally nothing in common with America’s.

  As Elbridge Durbrow, the State Department’s expert on the Soviet economy, put it, Stalin had just said, “to hell with the rest of the world.”62

  *

  *1A code, strictly speaking, is a system in which an entire word is represented by an individual symbol or numeral; in a cipher, each plaintext letter is replaced by another letter (or other symbol).

  *2Morse code, which dated to the earliest days of telegraphy and radio, dominated the military, maritime, and commercial communications of the era. Morse code was not a “code” in the sense of a security measure, but a system for representing the letters of the alphabet; a radio operator manually pressed a key to pulse a radio transmitter on and off to produce the pattern of long and short beeps that stood for each letter. A Morse code signal took up much less radio bandwidth than a voice signal, was far less susceptible to jamming, interference, or misinterpretation, and required only the simplest radio gear to send and receive. (The last commercial Morse message was sent in 1999.) By the 1930s a growing portion of commercial and government radio traffic was being sent using radio teleprinter, which linked two automatic typewriter terminals over the airwaves, eliminating the need for manual keying and copying but requiring bulky and expensive specialized equipment.

  *3Rotor machines, of which the German Enigma (invented in 1923) is the most famous example, remained a mainstay of cryptography into the 1970s. Each wheel was electrically connected to its adjacent wheels via a series of contacts, one for each letter of the alphabet, that ran around its opposite faces. A maze of wires within each wheel interconnected the two faces, scrambling the identity of the letters in random order. As the wheels rotated, a new scrambling pattern continuously emerged. Thus the cipher letter A might stand for the plaintext letter G at one position, Q at the next, foiling the simple method of cracking a substitution cipher by counting the frequency of each letter in the cipher text and assuming that the most common one stands for a high-frequency letter such as E. A cipher that employs an unvarying substitution alphabet for the entire message is known as monoalphabetic; the continually varying substitutions produced by a rotor machine constitute a much more secure “polyalphabetic” cipher.

  *4The addition was modulo 10—that is, digit by digit, without carrying—so that the resulting number was always the same number of digits long; for example, the code group 7829 combined with the additive 9234 yielded the enciphered code group 6053.

  *5A welter of code names further concealed the actual target of the projects. Following its standard practice of giving the codes of each country a color designator, the Army initially referred to the Russian systems as Blue. The Navy used the code name Rattan, sequentially designating each system it had identified with the letter R followed by a hyphen and a number. In June 1945 the Navy changed the code name to Bourbon and switched all the R’s in the system descriptors to B’s.

  2

  Unbreakable Codes

  It took Igor Gouzenko, a twenty-six-year-old code clerk in the Soviet embassy in Ottawa, forty hours to find someone to defect to.

  It did not help that the short, tubby, ashen-faced man was so petrified, he was barely able to explain himself to the Canadian newspapermen and government officials he approached for help. The larger problem was that to believe what he was saying—that his country had been engaging in a massive act of perfidy against a wartime ally, running a vast espionage network that extended to officials in Canadian government departments and the British High Commission, scientists working on the Canadian-British atomic energy program, even a member o
f the Canadian Parliament—required standing the world of 1945 on its head. Mackenzie King, Canada’s long-serving prime minister, registered in his diary the shock he felt when later confronted with the truth of Gouzenko’s claims:

  I think of the Russian Embassy being only a few doors away and of them being a center of intrigue. During this period of war, while Canada has been helping Russia and doing all we can to foment Canadian-Russian friendship, there has been one branch of the Russian service that has been spying on [us]….The amazing thing is how many contacts have been successfully made with people in key positions in government and industrial circles.1

  Gouzenko’s key proof was the texts of 109 cables he had transmitted to Moscow for Colonel Nikolai Zabotin, the resident head of the GRU, Soviet military intelligence, for whom he had been working since his arrival in Ottawa in June 1943. Fearing he was about to be sent back to the Soviet Union, Gouzenko had been making preparations for months, selecting the cables that he felt sure would buy him and his pregnant wife and young son a permanent home in the West. On the night of September 5, 1945, Gouzenko stuffed the documents inside his shirt and slipped out of the embassy—only to find himself brushed off as he tried for the next twenty-four hours to get the editor of the Ottawa Journal, the minister of justice, the Crown Attorney’s office, and the Royal Canadian Mounted Police to take his story seriously.

  The following night, exhausted from a day of fruitless trudging from office to office in the late summer heat, knowing that the embassy by now would be sure to be alarmed by his absence and the missing documents, the Gouzenkos returned to their apartment, and their despair turned to panic when a few minutes later a furious pounding on the door began and a Russian voice—Gouzenko recognized it as Zabotin’s driver—called his name. The man eventually gave up and went away. But at around ten o’clock four men from the embassy returned, with an officer of the NKVD, the Soviet state security service, in charge. A neighbor had meanwhile offered to let the Gouzenkos stay with them for the evening, and from across the hall they heard the Russians break down the door of their apartment. The local police at last arrived, and after a brief, tense confrontation in which the Russians claimed diplomatic immunity and tried to order the police off what they declared was Soviet property, the NKVD officer led his men away into the night.2

  The next day things began to happen. The Mounties took the Gouzenkos into protective custody and moved them to a safe house, a small, isolated lakeside summer cabin ninety miles from Ottawa, where Igor Gouzenko’s months-long debriefing began.3

  His most explosive revelations had to do with Soviet penetration of the atomic bomb program. The cables directly implicated a dozen Canadian scientists who had turned over to Zabotin myriad technical details gleaned from their American and British counterparts. Chief among the scientists was the physicist Alan Nunn May, who was part of a team building a large heavy-water reactor on the Chalk River north of the Canadian capital. Nunn May also served on two Canadian government committees that gave him access to high-level atomic secrets. One of Gouzenko’s cables consisted of a long report written by Nunn May at Zabotin’s request that described the entire organization of the Manhattan Project, including the work being done to produce weapons-grade uranium-235 and plutonium at Oak Ridge, Tennessee, and Hanford, Washington, and the names of the scientists who headed each section. Another cable revealed that Nunn May had given the Russians a ten-page, single-spaced typewritten technical report on the bomb project plus a microgram sample of uranium-233, a rarer isotope that the U.S. scientists believed might offer an easier pathway to producing weapons material from reactors.4

  In February 1946, still fearful of provoking a breach in relations with the Russians at a critical moment but feeling forced to act after the popular American newspaper columnist Drew Pearson, in his radio broadcast on February 3, broke the story of Gouzenko’s defection, Prime Minister King authorized the arrest of Nunn May and some twenty others implicated in the cables. Like nearly all of the U.S., British, and Canadian scientists who would soon be revealed to have engaged in espionage for the Russians, Nunn May was an idealistic Communist who was convinced that he was advancing the cause of world peace by sharing U.S. atomic secrets with the Soviets. “The whole affair was extremely painful to me,” he would later explain, “and I only embarked on it because I felt this was a contribution I could make to the safety of mankind. I certainly did not do it for gain.”5

  —

  An early visitor to Gouzenko’s lakeside cabin was Frank Rowlett.

  In response to a prompt entreaty from the British, the Canadians agreed to allow an American expert on “crypto matters” to come and interview the Soviet defector, and Rowlett left Washington with two days’ notice on September 25. He was driven out to the secret location accompanied by an RCMP inspector and a Canadian mathematician, Professor Gilbert Robinson, who had worked on signals intelligence during the war and had conducted a preliminary interview of Gouzenko to find out what cryptologic information he might have.

  Rowlett returned without any breakthroughs to report but with a significant haul of small details for the team at Arlington Hall working on the Soviet diplomatic codes. Gouzenko confirmed that the diplomatic traffic was enciphered entirely with one-time pads, each containing fifty five-number groups. The GRU’s messages, which Gouzenko exclusively handled, employed a different codebook from the ZET trade and ZDJ diplomatic systems that Arlington Hall had made the most progress on to date, but some of the details he was able to supply about the construction of the GRU codebook probably applied to those other systems as well.

  The GRU book was what was known as a one-part code: the words in the codebook were placed in alphabetical order and simply assigned their corresponding code numbers in ascending numerical sequence, from the beginning to the end of the alphabet. That permitted a single codebook to be used for both encoding and decoding; by contrast, a two-part code, in which numbers were assigned in random order to the words, required the preparation of two separate books, one arranged in alphabetical order for encoding, the other by numerical order for decoding. A one-part code eased some of the logistical problems for the codemakers, but it offered a great boon to the codebreakers, making it much easier to guess the meaning of an unknown code group by revealing where it fell in the alphabet relative to other already known groups.

  Another detail that Gouzenko supplied about the codebook was potentially even greater help. There was a special system, a code table within the code, used to spell out words using Latin or Cyrillic letters. This was needed whenever a foreign word, or a Russian word that had not been assigned its own four-digit code group, was transmitted. The special code group 7810 indicated “begin spell”; a two-digit group, 91, indicated the end of the spelled-out section. Within the spelled-out section, the numerical code groups stood for letters rather than words.

  If, as seemed likely, a similar spelling system was used in ZET and ZDJ, that would be a huge wedge to crack this traffic open. Like stereotyped openings, the “begin spell” group itself was likely to appear frequently in messages, offering an entry point. Moreover, any spelled-out section, once the additive was removed, was in effect a simple substitution cipher, in which the same numeral always stood for the same letter. Such simple ciphers are susceptible to the most basic of all cryptanalytic attacks, which exploit the fact that in every language some letters appear far more frequently than others, making it possible to guess the identity of the symbols that stand for each letter by counting how often each appears.*1 Once the spell table was cracked using this technique, the codebreakers would have a significant handle on reading depths and recovering additive, since it is often easy to guess the additional missing letters of a partially recovered spelled-out word.6

  The basic point was that if the codebreakers could make a reasonable guess as to what actual code group appeared at a particular spot in one message, then they immediately knew the value of the code group in the corresponding spot of a message in depth wit
h that one, since both messages had been enciphered with the same string of additive key. For example:

  message 1 2390

  – assumed code group 0234

  = recovered additive key 2166

  message 2 5987

  – recovered additive key 2166

  = new identified code group 3821

  Frequently occurring code groups that had been found this way could then be subtracted from every other message group to generate more “hypothetical key” that could be tested against other, unbroken messages to see if stripping this key yielded any other known, frequent code group, pointing to a possible depth.

  The Arlington Hall codebreakers meanwhile had made several other important discoveries. It had been known for a while that in the “Red” codebook used in the ZET trade system, the one thousand code groups that stood for numbers all followed a simple pattern. It was a common design feature in codebooks to employ a “garble check” for numerals to ensure that no mistakes had been made in encoding or decoding; this often involved having the code groups follow a redundant self-checking formula, such as 0102 for the number one, 0204 for two, 0306 for three, and so on. The trade messages clearly had little to offer of intelligence value—most were simply reports of goods being shipped under Lend-Lease—but for the same reason they were full of stereotyped content that made them a cryptanalytic bonanza, particularly when they contained number-laden lists of commodities and shipment quantities in rigid format.

  Samuel P. Chew, a professor of English at the University of Oklahoma who knew Ferdinand Coudert from when they were at graduate school together at Harvard, and who had taken the Army’s basic correspondence course in cryptology before the war, had made the crucial discovery about the stereotyped format of the trade messages in April 1945. Richard Hallock’s earlier key recovery work had been restricted to the first four or so groups of each message, where stereotyped openings were found, but Chew’s break made the full length of the one-time-pad pages potentially open to discovery by making it possible to guess the underlying code groups contained within the body of a message. Cracking the spell table would do the same for other messages as well.7