America's Spycraft

The Russian-Ukrainian conflict has been taxing to the American intelligence community; they have scored both achievements and failures. But to hear some of its exponents tell it, it has been an occasion for stupendous success. Learning of the false flag operations that Russia intended to use as a pretext for launching its war, the United States revealed the scheme to the world. As William Evanina, a former high-ranking counterintelligence official, tweeted: “Putin is fully aware the US Intel Community is the best in the world. Ever. The US collection is exquisite, deep, and without gaps, especially on Putin and his universe. He was not prepared, however, for how aggressive the US would be in sharing our collection to the world.”

Such boasting is out of place. It is questionable whether the United States has intelligence on the Russian leader that is “without gaps.” Only with a microphone surreptitiously planted on the clip of Putin’s tie could such a claim be true. There has been considerable uncertainty about Putin’s intentions after his initial hopes for a lightning victory were—unexpectedly to American intelligence—dashed by ferocious Ukrainian resistance. On the other hand, the American revelations about Russian false flag operations were indeed a master stroke in informational warfare. They were also, ultimately, a sign of impotence in the face of aggression.

To be sure, our limited military options have not been the fault of U.S. intelligence but something built into the situation of a nuclear superpower engaged in a conflict with a non-NATO nation state. Still, U.S. intelligence would be better off with a posture of humility than one of crowing. In the primary task of gauging threats against the United States, the eighteen agencies that today comprise the intelligence community have a long record of analytical and collection failures. The CIA missed the first Soviet atomic bomb test in 1949; the outbreak of the Korean war in 1950; the first Soviet H-bomb in 1953; the outbreak of the Suez war in 1956; the Soviet placement of missiles in Cuba in 1962; the Egyptian attack that started the Yom Kippur war in 1973; the Soviet invasion of Afghanistan in 1979; and the Iranian revolution that same year. As Amy Zegart points out in her new book, Spies, Lies, and Algorithms, in considering where next the United States will go to war, the track record has been far from perfect. Indeed, it has been perfectly wrong. “We have never once gotten it right,” she writes, “from the Mayaguez to Grenada, Panama, Somalia, the Balkans, Haiti, Kuwait, Iraq, and more—we had no idea a year before any of these missions that we would be so engaged.”

Zegart is a professor of political science at Stanford, a fellow at the Hoover Institution, and a longtime student of intelligence affairs. Her point (and mine) in highlighting this dismal track record is not to trash American intelligence. Rather, it is to indicate the intrinsic difficulty of making predictions in an unpredictable world. As we have moved into the digital era, and new tools have become available for understanding—and for carrying out deception—has the task become easier or more difficult? That is one of the questions implicitly raised by Zegart’s volume, a survey that ranges widely over U.S. intelligence to consider its history, its organizational structure, and its primary challenges.

As far as organization is concerned, the most striking aspect of American intelligence is its fragmentation. This has been a perennial source of trouble. In 1775, George Washington ordered a raid on a munitions depot in Bermuda. The ships sailed only to discover on arrival that another American unit, unbeknownst to Washington, had beaten them to the punch. A similar failure of coordination presaged the intelligence disaster of September 11. In part, as Zegart notes, difficulties connecting the dots are inevitable: “[G]etting the right information into the right hands at the right time is hard.”

The problem arises in part because of specialization. A tank commander on the battlefield has a need for a very different kind of intelligence than does a Secretary of Defense or a president. Different kinds of methods and organizations are required to satisfy the variety of demands. The problem is further compounded by the vast numbers of individuals who have access to secrets. In 2017, some four million people held security clearances, with 1.3 million enjoying access to “top-secret” information. Given how sprawling the intelligence world is, and how secretly it must operate, plenty of important stuff is going to fall between the cracks.

Some of those cracks have widened lately. Zegart offers a fascinating tour of the technological developments that are reshaping the intelligence world, such as the rise of private operators. Satellite reconnaissance used to be the exclusive preserve of major states. Today, the technology has radically improved and the price tag has plummeted, opening up the field to numerous entities. The first Corona satellites launched in 1960 offered photographs with a resolution of twelve meters. By the 1990s, commercial satellites were offering sub-ten meter resolution. Today, one finds commercial outfits offering photographs of thirty-one centimeters (approximately one foot) resolution. At the same time, the cost of a single frame has dropped from $4,000 to $10. Google Earth, a fine intelligence tool, is free. For good and for ill, anyone with a home computer and a small budget can gain access to breathtakingly accurate images of far corners of the globe.

The bright side of this easy access is the emergence of think tanks and research centers that are making serious use of satellite imagery to further our understanding of topics ranging from climate change to nuclear proliferation. In one example of many she adduces, Zegart points to collaboration between the University of Missouri and the National Geospatial Intelligence Agency involving the use of machine learning to identify surface-to-air missile sites in a wide swath of China. The resulting computer program was able to perform the task of analyzing satellite imagery eighty times faster than human analysts, scouring in a mere forty-two minutes an area three-quarters the size of North Korea.

But there are pitfalls. Not every private actor is competent. As Zegart puts it, “the landscape is filled with questionable data, shoddy analyses, pet theories, and political agendas.” An Iranian dissident group offered up false analysis of imagery, pointing to a secret nuclear facility in the basement of a Tehran-based corporation. The report was debunked within a week. A Georgetown professor together with his students studied an underground Chinese facility and came up with estimates of the size of the Chinese nuclear arsenal that were dramatically higher than official U.S. estimates. Congressional hearings and Defense Department heartburn ensued. This Georgetown study was eventually also debunked, but not before time and resources were wasted in uncovering its flaws.

Cyberwarfare is another technological realm that is new to the intelligence world. In 2007, the intelligence community’s annual threat report did not mention the word. Today, their report ranks cyber as a top-tier peril. It can come in many forms. Zegart ticks off five: stealing, spying, disrupting, destroying, and deceiving. We have already experienced cyberwarfare with Russian manipulation of our elections through troll farms that flood American social media with disinformation. And we have experienced a more kinetic kind as well, such as when the Colonial Pipeline, a critical artery supplying gasoline to the East Coast of the United States, was shut down in a ransomware attack in 2021, causing fuel shortages in many locations. This attack was carried out by Russian-based criminals, and it illustrates our tremendous vulnerability to keystroke warfare.

Zegart is particularly worried about “deepfakes,” digital impersonation technology that is developing rapidly, making it relatively easy to create images and videos of real people in unreal situations. Honest political candidates appearing to engage in dishonest transactions; artificially generated photographs of politicians in sexually compromising positions—the possibilities for mischief through deepfakes are endless. The design of this kind of software is ingeniously diabolical. As Zegart explains:

Deepfakes are growing more convincing and nearly impossible to detect, thanks to a breakthrough AI technique invented by Google engineer Ian Goodfellow in 2014. Called “generative adversarial networks,” the approach essentially pits two computer algorithms against each other. One learns to generate a realistic image of something while the other learns to decide whether the image is real or fake. Because these algorithms are designed to learn by competing, deepfake countermeasures are unlikely to work for long.

Countering deepfakes—and using them for our own intelligence purposes—is but one of dozens of new challenges facing the intelligence community. But along with all that is new, the traditional requirements of spycraft have not disappeared. It is still incumbent upon intelligence agencies to perform their basic task of gathering timely information from difficult-to-mine sources, and of making sense of that information for policymakers. Spies, Lies, and Algorithms, based upon thorough research, numerous interviews, and benefitting from Amy Zegart’s long-term immersion in the subject, covers the waterfront of the intelligence field, old and new. As America’s adversaries gather strength and act aggressively, one puts down her book with a deep sense of foreboding about what is to come.

Gabriel Schoenfeld, a contributing editor of American Purpose, is a senior fellow at the Niskanen Center.