One of the best-known corporate missions in history sounds like the oath of an idealistic librarian: to organize the world’s information and make it universally accessible and useful. This is a story about what happens when the word “universally” starts demanding footnotes.
In March 2010, Google made a move many saw as a rare act of corporate heroism. The company announced that it was no longer willing to comply with censorship requirements for Google.cn—the search engine tailored to mainland China. The formal trigger was cyberattacks targeting, among others, the email accounts of Chinese human rights activists. The real one was the accumulated tension between what Google promised the world and what it did in one particular market.
The decision looked dramatic: users of Google.cn were redirected to Google Hong Kong, where search results were not filtered to meet Beijing’s demands. The Western press wrote about principle. But there was a nuance that was easy to miss: Google did not leave China. Its advertising business continued to operate. Offices in Beijing and Shanghai stayed open. Investments in Chinese startups carried on as usual. Cloud services, the Android ecosystem, research labs—all of it remained. The departure concerned one specific product—the search engine—and even that departure was, essentially, a redirect.
For a company whose market capitalization rests heavily on its search business, giving up search in the country with the largest internet audience on the planet was painful. By 2017, China’s internet had more than 700 million users, and almost all of that space belonged to Baidu. Google looked at this market like someone who had stormed out of a restaurant, slamming the door, only to stand on the sidewalk across the street seven years later, studying the menu through the glass.
In spring 2017, work began inside Google on a project codenamed Dragonfly. It had other names too—Maotai and Longfei—which in itself says something about the scale: projects with one name usually fit in one room. The goal was to create a specialized Android search app for the Chinese market. An app that would operate by rules set by the Chinese Communist Party.
What did that mean in practice? According to internal documents that later reached journalists, the system was supposed to automatically exclude topics deemed forbidden from search results: human rights, democracy, protests, mentions of the BBC and Wikipedia. In place of removed results, users would see a disclaimer—a notice that some information was unavailable due to legal restrictions. Put bluntly, the search engine would honestly tell users it was lying—and keep lying.
What got me was the detail about weather and the environment. Dragonfly prototypes, according to the descriptions, substituted air-pollution and weather data with official Chinese figures. That is a level of adaptation that goes beyond simply filtering politically sensitive content. Air quality is information that determines whether you take your child outside for a walk. When a search engine starts substituting data on what people are breathing, the word “censorship” stops being enough.
Another element of the proposed architecture, described in leaked documents: linking search queries to users’ phone numbers. Precision matters here. Dragonfly was not a launched product. There is no public evidence that Google passed Chinese users’ queries to the authorities. This was a feature of a system still at the design stage—but that feature itself turned the hypothetical search engine into something fundamentally different from what Google offers in the rest of the world. A search engine in which every query has a first and last name is a dual-use tool, and the second use is obvious.
On August 1, 2018, The Intercept journalist Ryan Gallagher published the piece that moved Dragonfly from the category of corporate secret into that of public scandal. The documents he relied on were detailed enough to reconstruct the project’s architecture, and specific enough to rule out the interpretation that “this is just a research experiment.”
The reaction inside Google was sharper than management had probably expected. Company employees—engineers, researchers, managers—began signing internal letters demanding transparency. They wanted to know what their colleagues were working on, and they wanted the right to decide whether to take part. For Silicon Valley, where loyalty to an employer has historically been reinforced by stock options and free lunches, this was an unusual gesture. People were risking their careers over a question that had nothing to do with their salaries.
Among those who left was Jack Poulson, a senior Google research scientist. His resignation became one of the most visible public acts of protest. Poulson later sent a letter to the Senate Committee on Commerce, explaining why he felt unable to remain at a company designing a surveillance tool for an authoritarian state. He was not the only one to resign, but he became the face of internal resistance—in part because he was willing to speak about it publicly.
Google employees’ protests in 2018 were not limited to Dragonfly. At the same time, disputes were raging over the Pentagon contract known as Project Maven—an object-recognition system for military drones. The company found itself in an unusual position: criticized simultaneously for working with the U.S. military and for being willing to assist Chinese censorship. Two fronts, two utterly different accusations, both about the same thing: the line between “we make technology” and “we are responsible for how it is used.”
The Dragonfly scandal quickly outgrew the boundaries of corporate ethics and became political. Sixteen members of the U.S. House of Representatives demanded explanations from Google. Human rights organizations—Amnesty International, Human Rights Watch—issued harsh statements. To them, Dragonfly was a concrete tool that could be used to identify people seeking information about human rights in a country where such a search can have real consequences for freedom and life.
December 11, 2018—a date that deserves close attention to detail. Sundar Pichai, Google’s CEO, was testifying before the U.S. House Judiciary Committee. He was asked directly about Dragonfly. Pichai’s answer became a model of corporate linguistics: the company had no plans to launch a search product in China “right now” (right now). At the same time, he confirmed that internal development had taken place and named the size of the team involved—around one hundred people; leaks, meanwhile, spoke of nearly three hundred.
Two words—“right now”—did more to undermine trust than any leak could have done. There are no launch plans right now. But in six months? In a year? The phrasing was legally impeccable and communicatively disastrous. It confirmed that the project existed on a scale incompatible with the definition of a “research experiment”—a hundred or two engineers do not work on abstract exercises—and at the same time left the door open.
The context of the hearing matters too. Pichai was before the committee not because of Dragonfly as such—the agenda was broader, including questions about political bias in search results, data privacy, and market power. But it was the China project that became the point where abstract questions of corporate responsibility acquired the utmost specificity: behind the dispute over whether an algorithm ranked news sources properly stood a system designed so that the state could know which of its citizens were interested in democracy.
A separate thread in this story is the role of Sundar Pichai himself. According to published reports, the Dragonfly project was overseen at the CEO level. In December 2017, Pichai held a meeting with Wang Huning—one of the most influential members of the Chinese Communist Party’s Politburo Standing Committee, a man often called the party’s chief ideologue. Little is publicly known about what they discussed, but the very fact of a meeting at that level speaks to the seriousness of the intent.
Here it is useful to separate two questions that often merge into one in public discussion. The first is the filtering of search results—that is, content censorship. The second is state access to user data, legal requests, potential surveillance. These are related problems, but not identical ones. A search engine can filter results while not handing user data over to the state. And vice versa: a system can display unfiltered results while still giving authorities access to query logs. Dragonfly, judging by descriptions of the prototypes, was designed so that both problems were built into the architecture at once: both blacklists of topics and the linkage of queries to phone numbers.
For many years, Google management relied on an argument one might call the “doctrine of partial access”: it is better to give users some access to information than none at all. This argument was heard back in 2006, when Google.cn was just launching, and Eric Schmidt explained that being present in China with restricted search was better than being absent altogether. The logic is understandable and even possesses a certain honesty. But Dragonfly exposed its limits: when “partial access” includes substituting air-quality data and potentially identifying users through their searches, this is no longer a compromise.
By all available accounts, the Dragonfly project was shelved. There is no exact date of its official closure in the public record—typical of internal developments that were never announced and therefore cannot formally be “canceled.” In July 2019, Google vice president of public policy Karan Bhatia told a Senate hearing that work on Dragonfly had ended. That is the most definitive public confirmation.
I cannot shake the thought that the main outcome of this story was not the cancellation of one specific project, but the exposure of a mechanism. Dragonfly showed what decision-making inside a technology corporation looks like when commercial interest collides with declared values. Hundreds of engineers worked on a project most of their colleagues knew nothing about. Decisions were made at the CEO level. Ethical oversight—if it existed—did not prevent the development; the press leak played that role.
The question Dragonfly posed to the industry did not vanish along with the project. Any global technology company operating in countries under authoritarian rule faces the same fork in the road. Apple stores Chinese iCloud users’ data on servers managed by a state-owned company. Microsoft ran a censored version of LinkedIn in China (until its closure in 2021). Amazon provides cloud services through local partners. Each of these cases is its own point on the scale of compromise, and none of them came with a scandal on the scale of Dragonfly, though the ethical questions in each are no less acute.
The phrase “organize the world’s information” contains a quiet assumption: whoever does the organizing decides what belongs to the “world’s,” and what does not. In 2010, Google decided that censorship was incompatible with this mission. In 2017, Google began designing a system in which censorship would be a core function. In 2018, Google said there would be no launch “right now.” Between these three points lies not an evolution of position, but the swing of a pendulum, its amplitude determined not by principles but by pressure: market, political, reputational. The pendulum stopped only because all three forces happened to align at once—and no one can guarantee that they always will.