by DAVID KAHN
Not all the human communications that N.S.A. studies are coded. Into the headquarters building at Fort Meade come recordings of the cleartext chatter between Soviet pilots. An N.S.A. section transcribes these, not into ordinary Russian writing, but into a phonetic representation that retains the pronunciation variations of the speakers. These transcription sheets go to analysts in another section. They compare the pilots’ inflections with known dialectical pronunciations to determine where the men in a squadron come from. Long residence in one locality will sometimes shade an older pronunciation more toward the local one; the analysts can detect this and tell where the squadron is stationed. Slang and current phraseology assists in these determinations. When one pilot calls another “Ivan,” and Ivan replies, the characteristics of his speech are carefully noted in an enormous file with all other Ivans so that future clues can be fitted into the original ones to add more details. Jokes, comments about superior officers, references to nearby units, remarks about the planes, all are catalogued. Sometimes an analyst will spend days on a single sentence, checking and cross-checking names and intonations. And just as the tens of thousands of points of pure color that Georges Seurat dabbed individually onto his canvas combined into the huge and stately “Sunday on the Grand Jatte,” so the thousands of details elucidated by the analysts build up into a broad image of Soviet air power, fuzzier than the painting, of course, but with a great deal of collateral intelligence on capability, morale, equipment, and almost every subject which a potential adversary might find of interest.
But the National Security Agency produces its most valuable intelligence by breaking foreign codes and ciphers. And though “practical cryptanalysis” sometimes helps, most of the results come from true cryptanalysis. As Martin and Mitchell said: “Successes obtained by the National Security Agency in reading the code and cipher systems of other nations are due primarily to the skillfulness of cryptanalysts, frequently aided by electronic digital computers.”
Who are the cryptanalysts, and how many does N.S.A. have? It is difficult to give an exact answer, because modern cryptanalysis is so specialized and so subdivided that many N.S.A. employees engage in partial or elementary cryptanalysis, or do the nearly mechanical task of filling in the holes after the “real” cryptanalysts have made the entry into a code or cipher and have thoroughly broken it. However, a rough guess might place the number of “real” cryptanalysts in N.S.A.—those who attack unknown or new systems—at about 200.
Despite the great secrecy surrounding their work, and the great events that can flow from it, the cryptanalysts’ labors resemble those of any other office workers. At N.S.A., they arrive in one of three shifts, beginning at 7:20, 7:40, or 8:00 a.m. (and ending respectively at 3:50, 4:10, and 4:30 p.m.). Once in, the first order of business must be to finish reading the newspaper and shoot the breeze with one’s officemates. When they get down to work, they write on cross-ruled paper with colored pencils, shuffle pages, look for significant patterns, look for plaintext, confer with colleagues, take coffee breaks. Sometimes a yelp of joy will pierce the concentration as a cryptanalyst breaks through. They have one advantage at least over workers in more ordinary fields: they cannot take their work home with them at night. But, in another sense, they cannot get away from it, for a problem in cryptanalysis grips the mind, teases and torments it more than other problems, and never seems to let go. If an idea occurs at home, the cryptanalyst may write a note to himself, or, if he lives close enough, he might perhaps drive down to the headquarters building to work on it.
As in other large white-collar organizations, they probably work in large open offices. Into them come the raw intercepts—no doubt, in most cases, the typewritten copies as made on four-ply paper by the intercept operators. Urgent messages are most likely forwarded by radio, as the MAGIC intercepts were sent to Washington from the Philippines. If several versions of the same message, picked up by several intercept operators, reach Fort Meade, editors will try to clear any garbles. Presumably traffic analysts then collate and compare sender location, routings, and indicators. This enables them to sort the messages into families of identical cipher systems for the cryptanalysts. And by studying traffic patterns, they can deduce tables of military organization and perhaps other information as well.
The cryptanalysts work in teams. Complex modern ciphers have rendered individual work as much a thing of the past in cryptanalysis as in other branches of science. Thirty-three atomic physicists signed the report announcing the discovery of the omega-minus particle; seemingly as many N.S.A. cryptanalysts would deserve credit for solving the rotor system of a sophisticated modern nation.
The head of the team apparently parcels out such assignments as different statistical tests, calls conferences, decides whether one attack is proving more fruitful than another. The cryptanalysts’ work consists in essence of looking for textual patterns that deviate significantly from what could be expected by chance. These patterns are extremely tenuous, and the individual letters of which they are composed recur only at extremely long intervals. This results from the efforts of rotor systems, Hagelin machines, and computer-generated keys to make it as hard as possible for the cryptanalyst to assemble the monalphabetically enciphered letters that he must have to reach a solution. Only enormous quantities of text can make these faint patterns visible, and only huge data-processing computers can engorge the rivers of letters and test the innumerable possibilities to solve the system in real time, which is to say before it has lost its usefulness. For computer processing, key-punch operators very likely punch the messages on cards, and technicians feed the cards into the computers.
N.S.A. probably has more computer equipment than any other installation in the world. Some of those it reportedly has are general-purpose computers, such as the I.B.M. Stretch, one of the world’s fastest and most powerful computers, the $2,898,000 I.B.M. 7090, which can perform 229,000 additions per second, and late-model Univacs; there is also the Atlas, which N.S.A. had built to its own specifications early in the 1950s, and probably several smaller general-purpose computers. The agency also has a great deal of special computer equipment. For example, a device may be built to run the kappa test instead of wasting a general-purpose computer on so restricted a task. N.S.A. may use its computers to determine which configuration of possible displacements on a rotor produces the group of letters that most closely resembles plaintext. The giant calculators may solve or partially solve the equations of group theory needed in analyzing a rotor machine. They may run test decipherments, simulating rotors wired in various ways and turning in various periods, and print out the test solutions at rates up to 600 lines per minute, starring those solutions that statistically most resemble plaintext. Undoubtedly the agency has prepared and debugged programs for common routines and holds them in readiness for immediate use.
The computer has in no way conferred total victory upon cryptanalysis in its unending struggle with cryptography, for cryptography has kept pace with countervailing developments of its own. Nor has the computer automated the cryptanalyst out of a job. The computer has relieved him of much drudgery, but modern cryptosystems involve much more work than older ciphers. Computers could be programmed to recognize plaintext by stocking their memories with letter frequencies, 10,000 common words, and basic grammatical rules. But it could not do so as quickly as a human being. Furthermore, the computer would have to run through all of even the better possibilities in a modified “brute-force” attack—something which would take impossibly long. A human being can correct and enlarge partial solutions. And there is no machine yet devised that can, as quickly as the living computer inside the skull, make an inspired guess on the basis of a half-forgotten news item in the Washington Post of a month ago and last night’s television news that the formless mess of letters i-go-e-ia must be a garbled Indonesia. Finally, and above all, a human brain must decide which tests the computer should run on a sheaf of cryptograms. Cryptanalysis still has room—indeed, may have more room than
ever before—for flair, intuition, experience, individual brilliance. The computers at N.S.A. are—as they are wherever computers are used—the tools of their operators, not their replacements. They are robot cryptanalysts to a very limited degree. Thus, in the last half of the twentieth century, in the flowering of the computer age, cryptanalysis often comes down to exactly the same problem that four centuries earlier faced the West’s first great cryptanalyst, Giovanni Soro of Venice: Does x stand for a or o?
The quality of the systems N.S.A. attacks varies greatly from country to country. Competence in cryptology, as in other fields of endeavor, seems to vary in direct proportion to the technological knowledge and the economic wealth of a country. On this basis, the United States probably has the most secure cryptosystems and the most informative communications intelligence in the world. Of the nations whose cryptograms N.S.A. attempts to solve, unquestionably the most sophisticated must be the Soviet Union, Great Britain, and France, probably in that order.
In all probability, N.S.A. attempts to solve all cryptosystems of all countries—at least in principle. But manpower and monetary limitations afflict N.S.A. like other agencies, and these and the incessant emergencies that must require pulling a cryptanalyst off his regular task make the ideal unattainable. Thus, though N.S.A. might want to attack, for example, the middle-echelon military systems of a Near Eastern country, it might have to concentrate the cryptanalysts that would be assigned to it on a Russian system that could be expected to yield more valuable results. How long it will keep a team working upon a system probably depends upon the information it thinks it will obtain. The agency may well keep a team examining cryptograms in a given system for two or three years, even though it has had no success, in the hope that one of the cipher clerks may some day blunder and open the way to a solution. For in modern systems, properly used and with frequent key changes, a cryptographer’s error is the cryptanalyst’s only hope. And when nations will pay their code clerks only $60 a week, as Italy did in Washington in the 1960s, to await such errors may not be pointless.
In addition to the general cryptanalytic effort, N.S.A. may mount special attacks if one of its customers requests it. The State Department, for example, may request such a solution in advance of a high-ranking official’s visit to another country or before a major diplomatic conference.
N.S.A. cryptanalysts probably solve foreign cryptosystems in degrees of completeness that range from total reading of all messages in a given system, to fairly full solutions with a few questionable patches, to partial solutions with many holes, to solutions in which, say, one or two rotors of several have been reconstructed but no plaintext has been read, to an absolute blank. Solutions probably also vary in time: the cryptanalysts may read a complicated system for a few months, then lose out again in a change of key.
The solutions must go to organizations in the U.S. government that require that information—military details to the Defense Department, diplomatic to State, and so on. These, together with the C.I.A., must be N.S.A.’s chief customers. Probably each class of messages has a distribution list. Individual messages may well be read at meetings of the National Security Council and the U.S. Intelligence Board. During the Korean War, the White House itself reportedly called for solutions, even though some were fragmentary. Currently, the President sees the N.S.A. “Black Book” every morning, brought to him by his military aide.
What does it all consist of? How successful is N.S.A., and how valuable are its results?
It is likely that N.S.A. reads only a small minority of the total volume of intercepts sent it—perhaps under 10 per cent. In peacetime, encipherers can work more slowly and more accurately than in war—yet even in the wartime conditions of the Russian front, with a great volume of messages and unquestionably many more errors, Germany’s Army Group North solved less than 30 per cent of Russian military cryptograms. Moreover, the N.S.A. intercept posts probably concentrate on messages in the highest priority systems, yet these must be the best systems and must often resist solution, thus lowering N.S.A.’s average.
Nevertheless, N.S.A. does solve enough cryptograms to produce information of great value to the nation. Martin and Mitchell delineated the extent of N.S.A. success. The agency, they said, solved the codes of more than 40 nations—or just about half of all that there were when they spoke. Asked which ones, Martin replied: “Italy, Turkey, France, Yugoslavia, the United Arab Republic, Indonesia, Uruguay—that’s enough to give a general picture, I guess.” This range is remarkable. France is one of the world’s great powers and has a long and strong cryptologic tradition. It stands as an American ally in the free world, as do the other major European country (Italy), the small Latin American country (Uruguay), and the neighbor of Russia (Turkey). Indonesia and the U.A.R. are both important neutrals in the Cold War. Yugoslavia is a renegade Communist country. The two defectors would not say whether the United States reads Soviet messages. But the Soviet predilection for the one-time pad in diplomatic messages, and its known cryptologic sophistication, make it most unlikely, except by a lucky accident.
Hamilton, the Arab, filled in some details of the Martin-Mitchell outline:
I was listed as an expert on the Near East Sector in the office designated ALLO, which means “All other countries.” This sector concerns itself with the U.A.R., Syria, Iraq, Lebanon, Jordan, Saudi Arabia, Yemen, Libya, Morocco, Tunisia, Turkey, Iran, Greece, and Ethiopia. The duties of my colleagues in ALLO included the study and breaking of military ciphers of these countries, and also the deciphering of all correspondence reaching their diplomatic representatives in any part of the world…. N.S.A. reads the ciphers of all these countries by applying cryptanalysis….
I knew for a fact that the State Department and Defense Department systematically read, analyzed, and utilized in their own interests the enciphered correspondence between the U.A.R. embassies in Europe and the U.A.R. government in Cairo.
For example, I had in my desk all the deciphered communications between Cairo and the U.A.R. Embassy in Moscow relating to the visit of the U.A.R. government mission to the U.S.S.R. in 1958 for the purpose of purchasing petroleum in the Soviet Union. N.S.A. sent all these communications to the State Department just as it continually sends it the deciphered instructions of the U.A.R. Ministry of Foreign Affairs to its embassy in Washington….
It is especially important to note that American authorities take advantage of the fact that the U.N. headquarters is located on American soil. Their highhandedness has reached the point where the enciphered instructions of the governments of the U.A.R., Iraq, Jordan, Lebanon, Turkey, and Greece to their missions to the U.N. General Assembly fall into the hands of the State Department before arriving at their proper address.
The intelligence that flows out of Fort Meade mingles with intelligence from many other sources to help high officials determine national policy and tactics within the framework of American goals. N.S.A. intelligence is not as voluminous as C.I.A.’s, a former top C.I.A. official has said, but it is of a higher grade. All intelligence is evaluated for credibility, and cryptanalyzed intelligence must nearly always get the highest rating (some messages may be dummies) because it comes straight from the mouths of the subjects themselves. N.S.A.’s intelligence covers the gamut of communications of modern nations, from the minutiae of legation routine to the secret instructions to ambassadors. Even at its most complete, however, it can illuminate but part of the intelligence picture. The solutions allude to persons and facts and basic policies half known or unknown to the interceptor; they presuppose a common knowledge not at his disposal; they do not include information exchanged by personal contact, letter, telephone. Most messages mean little standing alone; only context makes them comprehensible. Cryptanalysis thus complements other forms of intelligence, overt and covert, just as they complement it.
Perhaps it is the incompleteness of cryptanalytic intelligence that led to American officials’ apparently disbelieving it at the time of the Suez crisis, de
spite its seemingly unimpeachable authenticity. Just after that crisis had passed its peak, George Wigg, a Labor Member of Parliament, told newspapermen that the United States had broken British, French, and Israeli codes and so had prior knowledge of plans for their invasion of Egypt at the end of October and beginning of November, 1956. Though he attributed the solution to the “United States Air Research and Development Command, Griffis Air Force Base, Rome, New York,” Wigg’s basic point seems to have been independently confirmed by C.I.A. chief Allen Dulles, who wrote several years later of the Suez invasion: “Here intelligence was well alerted as to both the possibility and later the probability of the actions taken by Israel and then by Britain and France.” Why, then, did the United States take no action? Dulles does not say, but Wigg thought “that the United States State Department knew from the middle of October what the French and the Israelis were planning to do. What I think they may have doubted was that the British Government would ever be so foolish as to get caught up in an adventure which was bound to end in disaster.” Secretary of State John Foster Dulles said at the time: “We had no advance information of any kind.” The later contradiction by his brother Allen suggests that this may be a cover-up for failure to act. Wigg, moreover, is not an M.P. whose inside information can be taken lightly: in 1963 he exposed the John Profumo-Christine Keeler scandal that very nearly toppled England’s Conservative government. Suez has been called one of America’s worst intelligence disasters. It seems more likely that the fault lay, not with the producers of intelligence, but with the consumers. No human being has ever had difficulty in finding an excuse to overlook an unpleasant fact. The consumers did not want to believe the contrary evidence of the cryptanalyzed intelligence (assuming that it existed). So they simply did not believe it—and perhaps justified their disbelief on the basis of its incompleteness. Against this human predilection no form of intelligence can prevail.