The Closure of the Internet

The Most Important Social Phenomenon of the Late 2010s — Arctotherium · Aug 09, 2026 — Not With a Bang

IntroductionIn July 2012, the UN Human Rights Council (UNHRC) unanimously adopted a landmark resolution affirming a simple proposition: “The same rights that people have offline must also be protected online, in particular freedom of expression, which is applicable regardless of frontiers and through any media of one’s choice.” Writing in The New York Times, Carl Bildt, Sweden’s foreign minister [at the time], hailed the vote as a “breakthrough of fundamental importance,” insisting that “the free flow of information on the Internet is a global call and not something pushed only by a few Western states.” For a brief moment, it seemed as if the world’s democracies had accepted that the open Internet would be the nervous system of modern liberty. (The End of the Open Internet) Until 2015, an optimistic and liberal (in the best sense) ethos pervaded the Internet. Major Internet platforms were open and had little content policy beyond bans on real-life criminal activity (stalking, selling illegal goods, child pornography, etc). Aside from these narrow and reasonable explicit exceptions, anyone could go online and converse with interested people around the world about whatever they wanted in public. These platforms had no real means to control what people talked about on them even if they wanted to, which they didn’t; to the extent an organization can have views all believed in free1 and open online discourse, which promised to make censorship impossible and end the hegemony of 20th century broadcast media.In this, they were backed up by major non-tech institutions, such as the UN, the EU, the US State Department, and the New York Times2, all of which saw Internet freedom as a corollary of offline freedom, and one which would liberalize the world’s remaining non-democratic states in time. These institutions both supported Internet freedom rhetorically3 and spent hundreds of millions of dollars creating and disseminating technical tools to evade Internet censorship and preserve online anonymity.The last gasp of the former left-wing institutional view that freedom of the Internet was a good thing might have been New York Times journalist Zeynep Tufekci’s Twitter and Tear Gas, which I read when it was published back in 2017.But in just a few years, every single major platform had created strict, but incredibly vague and opaque, ideological content rules backed up by a colossal moderation/censorship apparatus with tens of thousands of human moderators and an entire new field of machine learning created from nothing. The Internet went from a space for people without institutional backing to get their views out to one with regular purges and demonetizations of heterodox figures and those associated with them, encouraged by those very same non-tech organizations that formerly championed Internet freedom. This was justified as a response to (massively overblown and mostly fictitious) Russian influence campaigns4, and as fighting nebulous “hate,” the definition of which could be shifted at will to cover whatever views or ideas the organizers classifying it wanted it too and to exclude those they didn’t, and “misinformation.”The closure of the Internet occurred in two distinct phases: 2015-2019 (associated with the 2015 migrant crisis, Brexit, Donald Trump’s election, and the so-called alt-right) and 2020-2022 (associated with the lockdowns and the 2020 election). Almost all of the technical and organizational infrastructure and conceptual justifications were built out in the first phase, which successfully decapitated and destroyed the previously-independent, but by definition fringe, online ecosystem of ideas. In the second phase, the same weapons that previously targeted heterodox online discourse were turned against opposition to the COVID measures du joure and mainstream conservatives. The scale of censorship has no precedent; tens of millions of accounts suppressed or banned, billions of posts removed. These purges are usually framed as a direct response to particular events or atrocities. For example Amazon ostensibly removed all Confederate flag merchandise from their web store in response to Dylan Roof’s 2015 massacre of black churchgoers. This is a mistake. Atrocities without a propaganda campaign from the press and political pressure from an opaque network of governments and quasi-government NGOs lead to no changes. There were no mass purges after Audrey Hale’s 2023 murder of elementary schoolers in Nashville motivated by anti-white hatred or the 2016 murder of five Dallas police officers by an ex-marine motivated by the BLM movement. On the other hand, press campaigns not centered around an atrocity, such as the mass deplatformings of YouTube personalities, lead to the same types of purges as the atrocity-based ones. It is not the atrocity that causes the changes, but the media and behind-the-scenes pressure campaign.The late 2010s also featured suppression of heterodoxy offline as well. This combined physical attacks on the streets, as happened to Charles Murray, and the firing of public figures with non-Internet platforms, as with Bo Winegard and Bryan Pesta. Even moderates and liberals in already left-wing institutions were not immune, as the purges of those who valued pluralism and objectivity within the New York Times, ACLU, and many progressive activist organizations showed. As such, the closure of the Internet shut down all avenues for heterodox discourse during the Great Awokening. If you spoke from an institutional platform, you would be fired. If you spoke in person, you would be beaten or shot. And if you spoke online, you would be censored, suppressed, demonetized, and doxxed.The closure of the Internet is the most important social phenomenon of the late 2010s. Most young people live online. More than half of American adults under 50 say they use the Internet “almost constantly”. Globally, almost 70% of people are online, and the average Internet user spends almost seven hours a day there. Cyberspace is where most couples meet, where most friend groups hang out, and where most people have fun. There is no alternative5 for even semi-serious intellectual or hobbyist discussion outside of a handful of low-capacity and carefully gatekept institutions. The word “censorship” does not go far enough; online went from almost completely open and free to a series of “closely monitored playpens,” like going from the American frontier to East Germany in just a few years6.EscalationFirst they came for the Nazis, and I did not speak out—Because I was not a Nazi.The clearest pattern across all of these platforms was one of very rapid escalation. The closure of the Internet is one of the clearest examples of the slippery slope (or “no limiting principle”) in action; it only took a few years to go from deplatforming the fringe neo-Nazi website Stormfront to the entirely mainstream New York Post, from banning books on Holocaust denial to banning those opposing transgender maximalism, and from banning honestly discussing the effects of immigration in Europe to banning honestly discussing the lab leak hypothesis of COVID origins. It is one thing to defend freedom of speech or freedom to create a website in the abstract and quite another to defend Andrew Anglin calling to exterminate the Jews. And even the most unpopular political or intellectual expression is easier to stomach than something like r/watchpeopledie. But the free speech maximalists were right. Once the conceptual and technical architecture for identifying, deplatforming, suppressing, and censoring “hate” and “disinformation” was built, the institutions defining these terms quickly extended them to as many inconvenient facts, ideas, and positions as they could.Conceptual UniverseA whole new conceptual universe was constructed to justify and organize the closure of the Internet. For example, the words “disinformation” (false information shared maliciously), “misinformation” (false information7 shared inadvertently), and “malinformation” (information that is true but misleading) practically did not exist until 2016, when in response to Trump’s election they were rustled into existence by a coterie of governments, academics, and nonprofits.Source: David Rozado.The meaning of the term “online safety” shifted from concerns about people using the Internet to physically attack, stalk, and harass people in real life to horror at the idea of users seeing marginalized ideas and information. This led to the Orwellian naming of so-called “Online Safety Acts,” which rather than focus on preserving anonymity and pseudonymity, which protect the physical safety of people online, instead work to eliminate them. A similar definition shift occurred with “harassment and abuse” and “violence.” This reached its apogee with the slogan “silence is violence,” the totalitarian idea that one must not simply not oppose orthodoxy but actively and enthusiastically propagate it or be treated as a murderer and therefore legitimate target.The Soviet concept8 of “hate speech9”, first imported at scale to the West during the 60s revolution, went from fringe to a common term in everyday discussion.“Hate speech” was an alien concept to the Anglosphere until the 1960s, and remained fringe until the closure of the Internet. Source.Relatedly, the definition of the word “toxic” was changed from a synonym of “poison” to a descriptor of to-be-marginalized, but not illegal or against formal platform policies, online discourse, and an entire subfield of natural language processing, automated moderation/suppression of “toxic” discourse spun up from almost nothing in 18 months in 2016 and 2017.The meteoric rise of the terms “toxic comment,” “toxic speech,” and “toxic language detection” (for the machine learning field created to suppress them), after 2015. Source.Until mid-2016, this field of machine learning, “toxic language detection,” did not exist.This chart is a reproduction of figure 12 from “A Systematic Bibliometric Analysis of Hate Speech Detection on Social Media Sites” (2022), with the year 2021 excluded due to incomplete data and quality increased, showing the rise of machine-learning papers on hate speech detection on social media using the Scopus corpus of academic research. The shape of the chart— showing that this field effectively did not exist until 2016, is more important than the precise numbers— which will vary depending on exactly how you measure.The first big foundational dataset in the field (using “a list of criteria founded in critical race theory”) was published in June 2016 by Zeerak Waseem and Dirk Hovy. Along with academia, Google Jigsaw spearheaded this creation of this field, publishing a much larger Wikimedia dataset in October 2016 and launching Perspective API as a baseline for the rest of the moderation/censorship industry (“empowering moderators to identify and address toxicity in comments”) to build on in February 2017. The first academic workshop on “Abusive Language Online” was held in August 2017. By December 2017, the field was fully institutionalized10, with Kaggle and Jigsaw launching the “Toxic Comment Classification Challenge,” which received over 4000 entries.The last major piece of this conceptual universe was the word “platforming.” This word shifts the burden of action; rather than exchange of ideas being the default and censorship something that must be justified, censorship becomes the default and permission to speak the exception that must be argued for. A person or group or idea is no longer “censored” or “banned” but “platformed.”Platforming is also a technical term used in engineering. This chart thus understates the true rise in the concept after 2013. Source.Platform-by-PlatformThe closure of the Internet can be described in broad strokes, but it is worth tracing the specific dynamics on each of the major platforms circa 2015, when it began in earnest. Every major platform that existed in 2015 went through th