A few weeks after Elon Musk formally acquired Twitter in October 2022, a senior official at the company who quit in the wake of Musk’s arrival took to the New York Times to pour cold water on Musk’s vision for the social-media platform. Yoel Roth, whose title had been Head of Trust and Safety, sought to assure his fellow progressives. Roth wrote that even if Musk wanted to remove the web of content-moderation rules and procedures Roth had helped create and enforce, the tech billionaire would be unable to achieve his aim. “The moderating influences of advertisers, regulators and, most critically of all, app stores may be welcome for those of us hoping to avoid an escalation in the volume of dangerous speech online,” he wrote.
What Roth meant was this: No Internet platform is an island, and Musk simply didn’t have the power to do what he wanted despite his 100 percent ownership of the social-media platform. It wasn’t merely that Musk would have to contend with Twitter’s progressive workforce, which believes that some political speech is so awful that it should be throttled or banned. He would also come into conflict with European regulators, the Federal Trade Commission, and Congress, all of whom also seek to limit what can be said online. And what about the Global Alliance for Responsible Media, a trade organization of some of the world’s biggest consumer brands that advocates for “online safety”—a euphemism for protecting social-media users from accounts that may offend, harass, or trigger them?
He would also be dogged by advocacy groups such as the Southern Poverty Law Center and the Anti-Defamation League, which have found a new and lucrative mission monitoring social-media platforms for hate speech. They work hand in hand with elite journalists and think tankers, who have taken to tracking the spread of misinformation and disinformation online. In Washington, the FBI and the Department of Homeland Security have personnel whose job it is to alert social-media companies to foreign propaganda and terrorism. In Atlanta, the Centers for Disease Control seeks to quarantine dangerous information that might lead Americans to forgo masks or vaccine boosters. And perhaps most important, there are other Silicon Valley giants—Apple and Google—that provide the digital storefronts or app stores that Twitter needs to update their software and continue to run its service.
Call it the “content-moderation industrial complex.” In just a few short years, this nomenklatura has come to constitute an implicit ruling class on the Internet, one that collectively determines what information and news sources the rest of us should see on major platforms. Talk about “free speech” and “the First Amendment” may actually be beside the point here. The Twitter that Musk bought was part of a larger machine—one that attempts to shape conversations online by deamplifying, muzzling, and occasionally banning participants who run afoul of its dogma.
The existence of this nomenklatura has been known for a few years. But thanks to Musk and his decision to make Twitter’s internal communications and policies available to journalists Matt Taibbi, Bari Weiss, and others, more detail is now known on why and how this elite endeavors to protect us from all manner of wrongthink.
The unspooling of Twitter’s secrets—in a series of long Twitter threads—has been revelatory for many reasons. It turns out that despite armies of human content moderators, artificial-intelligence tools to weed out everything from health misinformation to hate speech, and an expanding set of internal rules, a handful of senior executives still made the most consequential decisions on what Twitter’s users were allowed to see.
Twitter has taken great pains to appear as a neutral arbiter of allowable speech on its platforms. It has released semi-annual “transparency reports” for the last decade, providing data on the total number of accounts it has suspended and how many requests from governments it abided by. Along with Google, Meta, and others, Twitter signed onto the Santa Clara Principles on Transparency and Accountability in Content Moderation, a declaration of rights of social-media users to due process and free expression.
And yet the process at Twitter for moderating content from the largest and most controversial accounts was often arbitrary, ill-informed, and biased in favor of the beliefs of coastal progressives. Take, for example, the ban on sharing the New York Post story about Hunter Biden’s laptop in the run-up to the 2020 election. At the time, Twitter’s decision was presented as enforcement of its policy not to allow hacked materials to be posted on the platform (the laptop of the then–presidential contender’s son had been abandoned at a Delaware computer shop the year before). But as the first installment of the #TwitterFiles showed, no one at the company knew for sure whether the laptop included actual hacked material. In one memorable exchange, Twitter deputy counsel James Baker (former general counsel of the FBI) acknowledged as much but then opted for blocking the story out of caution. The decision was particularly glaring in light of Twitter’s refusal to block a New York Times story a few weeks earlier that published some of Donald Trump’s non-public tax returns—material that had unquestionably been purloined.
This kind of inconsistency proved to be a theme. Determinations that rules and procedures on the platform had been violated were often little more than judgment calls. Senior executives actually acknowledged that Trump’s last tweets hadn’t violated its policies when his account was permanently banned two days after the January 6 riot. LibsOfTikTok, an account with more than a million followers that usually posts excerpts of videos of deranged Millennials spouting the latest pabulum of gender theory, was suspended six times in 2022 for violating Twitter’s rules against hateful conduct. And yet an assessment for the senior executive committee found that LibsOfTikTok “has not directly engaged in behavior violative of the Hateful Conduct policy.” Rather, the assessment said the account had been harassing hospitals and medical providers by claiming that aiding gender transitions for adolescents amounted to “child abuse or grooming.”
The #TwitterFiles also exposed how the rules of content moderation themselves were subject to wildly different interpretations. Take this tweet from former president Donald Trump after he was released from Walter Reed Hospital where he had been treated for Covid: “Feeling really good! Don’t be afraid of COVID. Don’t let it dominate your life.” Baker asked why Trump’s tweet didn’t violate Twitter’s policy on Covid-19 misinformation, “especially the ‘don’t be afraid of COVID’ statement.” Roth responded that vague statements of optimism were not “misinformation.”
One of the biggest takeaways from the #TwitterFiles is that the FBI and the U.S. intelligence community were playing a significant role in content moderation. As Taibbi reported, several U.S. government agencies, and state and local law enforcement, used the bureau’s liaison portal with Twitter to flag more tweets than anyone previously suspected. Some of the tweets were from low-follower accounts whose proprietors were making dad jokes about when the election was scheduled. One Twitter content moderator openly expressed his fear that FBI agents were doing random searches for the specific purpose of finding violations of Twitter’s rules.
The American public has been led to believe that the FBI’s relationship with social-media companies after the 2016 election was designed to help foil and expose foreign propaganda. But the #TwitterFiles show that the FBI’s San Francisco bureau and the bureau’s Foreign Influence Task Force became a general conduit for requests from local law enforcement, the U.S. intelligence community, and others to take action against accounts with little to no proof of a foreign connection. What had begun as a response to Russian meddling in the 2016 campaign had become by 2020 an open door for the federal government to act as Twitter’s partner in content moderation.
One obvious question raised by the Twitter files is how social-media companies allowed the new Internet nomenklatura to dictate the terms of content moderation in the first place. Let’s examine this.
In the 1990s, during the first gold rush in Silicon Valley, the politics surrounding the concept of free speech were radically different from what they are today. The right was concerned about the dangers of offensive speech; the left wanted to leave the marketplace of ideas untouched. Just as liberals and conservatives tussled over Howard Stern and gangster rap, they also crossed swords over the Internet.
At first, social conservatives won the day. In 1996, Congress passed a law known as the Communications Decency Act, which, among other things criminalized “patently offensive” or “indecent” content on the Internet if it was plausible that adolescents and children could view it. This was the culmination of a debate that had been raging in America since the 1960s over whether local communities could prohibit a neighborhood newsstand or bookstore from selling pornography. The right largely lost these challenges. But at the height of the Clinton years and at the dawn of the Internet, the argument that unprecedented dangers were posed by the unlimited capacity of the Internet to disseminate information briefly triumphed.
Not for long. In 1997, the Supreme Court struck down most of the Communications Decency Act on First Amendment grounds. What remained was a compromise crafted by Representative Chris Cox and Senator Ron Wyden. Internet service providers would be encouraged to enforce terms of service to limit the most offensive material from being posted. In turn, according to the language of the new bill’s section 230, “no provider or user of an interactive computer service shall be treated as the publisher or speaker of any information provided by another information content provider.” This formulation—which meant that Internet platforms were to be treated pretty much as though they were the new era’s telephone wires and exchanges rather than purveyors of the material posted on them—made the Internet as we know it possible. Everything from Internet pornography to the comment sections on news sites to Facebook, Twitter, and Instagram owes a debt to section 230.
There was an important exception. Internet Service Providers still had to cooperate with law enforcement on criminal cases, such as those against child pornographers. These providers also had to police their servers for copyright infringement. This meant that content moderation in the late 1990s was mainly focused on taking down Napster and other sites that allowed people to acquire a digital library of music for free. Hate speech, harassment, doxxing, disinformation, conspiracy theories, and foreign propaganda—these were all fine as the world began to get online.
The protection provided by section 230 was a license for disruption. And the first social-media companies took full advantage. Just look at some of the mottos associated with the founders of social media. Facebook in its early years adhered to the creed, “move fast and break things.” Twitter executives once described their company as “the free speech wing of the free speech party.”
This anarcho-libertarian approach to content moderation is captured in a vignette from Pyra Labs, the company started by Twitter co-founder Ev Williams in 1999. He invented the software known as Blogger, which, along with Kinja and LiveJournal, provided the code that started the blog revolution in the early 2000s. The mantra for Blogger was “push button publishing for the people,” and it meant that anyone could publish what he wanted on the Web.
Nick Bilton’s 2013 book, Hatching Twitter, features a scene in which a new Pyra employee is confounded by a customer-service email complaining about a blog that displayed an animated picture of a group of men having sex on a trampoline. Asked what he should do about it, Williams replied, “Nothing.” Bilton writes that Williams soon “realized it would be impossible to police all of the posts that were shared on the site, so as a rule, he opted for an anything-goes mentality.”
This “anything goes” approach to content moderation carried over to the next venture for Williams, Twitter. He co-founded the platform with Jack Dorsey, Biz Stone, and Noah Glass in 2006. For its first years, Twitter really was the Wild West of social media. Trolls created accounts impersonating other users. Users could be doxxed—meaning their home addresses and other personal information could be published. Ethnic and racial slurs were allowed. And for a while, none of it seemed to matter. Twitter was a breakout hit in Silicon Valley. It was making no money, but by 2009 it was valued at more than $4 billion. Celebrities flocked to the platform. In 2009, Ashton Kutcher challenged CNN to a race for who could get to a million followers first (Kutcher won).
In 2011, Stone published a blog post on behalf of the company with the title: “The Tweets Must Flow.” He explained that there were times when the company would have to remove tweets from the platform, such as with spam or tweets that violated local laws. But, he wrote, “we make efforts to keep these exceptions narrow so they may serve to prove a broader and more important rule—we strive not to remove Tweets on the basis of their content.”
In this period, Twitter, Facebook, and other social-media platforms also strived to be politically neutral. For example, during the 2009 Iranian uprising over the stolen presidential election, the State Department persuaded Twitter to delay a long-scheduled maintenance update that would have disabled the service while protestors were planning a major demonstration. Twitter agreed, but after the news leaked that it had done so at the request of the U.S. government, Stone and Williams became worried. “Now it seemed Twitter had been seen as picking a side in an international war of words,” Bilton wrote. “It had been seen on one side of a moral and diplomatic fence—exactly the last place it wanted to be.”
An anecdote from Twitter’s larger rival, Facebook, also demonstrates the seriousness of social-media companies’ initial efforts to stay out of domestic politics. In 2016, on the tech-news site Gizmodo, a former curator of Facebook’s “trending topics” blew the whistle on the operation. This anonymous contractor described a kind of Facebook newsroom where Ivy League–educated contractors chose news items to feature next to Facebook’s newsfeed. The whistleblower claimed that almost all the curators were liberal and that news favorable to the right was suppressed. Fearing a backlash, Facebook decided to fire the contractors and leave the work of selecting news stories for “trending topics” to an algorithm.
This was a textbook case of creating a new problem to solve an old one. The algorithmic solution to the perception of bias at Facebook was jet fuel for fake news. Junky websites that trafficked in outrageous clickbait were favored by algorithms that sought to increase engagement. In his book Facebook: The Inside Story, Steven Levy writes that without the human oversight for “trending topics,” the algorithm “rewarded the types of posts that thrived on the News Feed—attention-getters, without regard to truth, good intentions, or newsworthiness.” Almost overnight, “trending topics” spotlighted stories from fraudulent sites, such as the “Denver Guardian.” Items would claim that Hillary Clinton had died or that the pope had endorsed Donald Trump. In this moment, Facebook was still moving fast and breaking things. After Trump was elected, though, Facebook would slow down.
There is a common misconception that the heavy-handed content moderation at Twitter, Facebook, and other social-media sites really began after the 2016 election—when the social-media companies, Congress, and the FBI discovered that Russia had been using these platforms to spread disinformation and misinformation.
That is true to an extent, but the ground was prepared for this before Trump. In 2015, for example, Twitter finally took on the issue of harassment. For the first nine years on the site, Twitter handled reports of abuse on an ad-hoc basis. Users were encouraged to report offensive tweets, and moderators would decide whether they violated the rules. Twitter allowed you to block abusive accounts, but that was about the extent of their efforts at what is today known as “online safety.”
Then This American Life, an NPR show, released an episode in February 2015 by feminist writer Lindy West about an anonymous troll who had created an account based on the identity of her father. Paul West had passed away only 18 months earlier from prostate cancer. The account featured a photo of West and a bio that read “embarrassed father of an idiot.” The piece was powerful enough that Twitter employees raised it in the company’s internal message system. In response, the CEO at the time, Dick Costolo acknowledged, “We suck at dealing with abuse and trolls on the platform and we’ve sucked at it for years.”
Two months later, Twitter’s general counsel at the time, Vijaya Gadde, wrote a mea culpa in the Washington Post that attempted to straddle the line between the company’s founding ideal of “letting the tweets flow,” and a new ethos of protecting vulnerable communities from online harassment. “Freedom of expression means little as our underlying philosophy if we continue to allow voices to be silenced because they are afraid to speak up,” she wrote. Gadde pledged that Twitter had tripled the size of the team devoted to online safety and predicted Twitter would be able to respond to user complaints in a fraction of the time it had taken before these changes.
Around the same time, Facebook was also confronting a similar problem but on a much larger scale. In the early 2010s, Facebook expanded its services all over the world, including countries with no tradition of free speech or digital literacy. At the time, this was all seen through the prism of the Iranian protests and the Arab Spring. Social media still had a halo around it. Facebook was a company that was bringing the world closer together. It was also a company that was enabling a genocide in Burma.
That was the conclusion of a blistering 2018 UN Human Rights Council report on the ethnic cleansing campaign against the Muslim Rohingya minority in Burma. “Facebook has been a useful instrument for those seeking to spread hate, in a context where, for most users, Facebook is the Internet,” the report said. “Although improved in recent months, the response of Facebook has been slow and ineffective.” The campaign had gone back years. On June 1, 2012, according to Levy’s book, the Burmese president’s spokesperson took to Facebook to call for action against the Rohingya. Levy wrote that the post essentially generated “support for a government massacre that would indeed occur a week later.”
Another factor in the 2010s that shifted the balance on the Internet from free speech to online safety was a particularly effective English-speaking, American-born jihadist named Anwar al-Awlaki. His video sermons were widely distributed throughout the Internet, and they helped to radicalize Nidal Hasan, the U.S. Army psychologist who murdered 13 people at Fort Hood in 2009. Several other followers of al-Awlaki engaged in terrorism at his urging, sometimes in personal correspondence. In 2011, President Barack Obama ordered his killing in a targeted drone strike in Yemen. But the videos of his sermons survived. For the next seven years, it was still relatively easy to find them on YouTube and other places on the Internet. It was not until 2017, that Alphabet (formerly Google), the parent company of YouTube, removed his videos from the platform. Until then, its own content-moderation rules allowed videos from al-Awlaki so long as they did not directly incite violence.
Part of the reason YouTube removed those videos is that lobbying from the U.S. government, academics, and pressure groups focused on online radicalization. Another factor was the rise of ISIS, which used Twitter, Facebook, and other social-media platforms to recruit and spread its propaganda on a much larger scale than al-Qaeda did. A Twitter blogpost from early 2016, for example, says the company suspended over 125,000 accounts for threatening or promoting terrorist acts in the last half of 2015 alone.
After Donald Trump unexpectedly won the 2016 election, the social-media platforms became a scapegoat for those seeking to explain Hillary Clinton’s gobsmacking loss. Facebook came in for a beating in Congress from Democrats who were livid to learn that a Kremlin troll farm known as the Internet Research Agency had purchased ads and created memes that denigrated Clinton.
There were two major problems for Facebook during the 2016 election that had prevented it from catching the Russians in the act. To begin with, Facebook was so focused on increasing its advertising revenue overseas that it didn’t put the ads themselves under much scrutiny at all. The second problem was that the Russian memes and posts on their platforms were not in any real violation of Facebook’s terms of service. Because most of the posts were about the 2016 election, the content fell well within acceptable political speech, according to Facebook’s own rules. When Facebook in 2017 began to suspend Russian accounts, it did so because they were fakes—the Russians were pretending to be people they were not—not because their content was unacceptable.
Twitter was also blamed for the 2016 election. As Taibbi reported on January 3, the company’s first audit for Russian activity revealed next to nothing. But as outside academics, reporters, and Congress began to claim otherwise, Twitter came to accept the narrative. “Researchers took low-engagement, ‘spammy’ accounts with vague indicia pointing to Russia (for instance, retweet activity),” Taibbi wrote, “and identified them as not only Russian, but specifically as creations of the media’s favorite villain, the Internet Research Agency.”
The focus on Russian meddling created an environment where it was possible for the FBI and other government agencies to play a persistent and direct role in content moderation for Twitter. By the 2020 election, what had begun with discreet requests from the U.S. government or committees in Congress to suspend accounts suspected to be Russian fronts transformed into a firehose of diktats from federal, state, and local authorities to censor content. Twitter was no longer a disruptive force for free speech; it was a tool for controlling discourse on the Internet.
This excessive discourse control often stymied the flow of real information in the name of filtering out disinformation. In his installment of the #TwitterFiles, which focused on the platform’s moderation of health misinformation, the gadfly critic David Zweig asked a devilish question. If Twitter had allowed dissent from the Covid dogma on its platform, would schools have opened sooner? Would the lockdowns have lingered?
Not every account that was suspended or throttled during the pandemic was giving sound medical advice or speaking truth to power. But plenty of them were. Zweig found that a tweet from Harvard epidemiologist Martin Kulldorff asserting that children and people with prior infections do not need vaccinations—a position in line with vaccine policies in other countries—had come into Twitter’s crosshairs. Kulldorff’s tweet deviated from recommendations of the Centers for Disease Control on vaccines. So Twitter labeled it as health “misinformation.”
One tweet alone would not have shortened the lockdowns or opened schools. But Zweig says that he found numerous examples of tweets taken down or labeled simply because they challenged whatever the CDC policies were at the time. Add to this Twitter’s efforts to shadow-ban accounts that countered conventional Covid wisdom. Weiss disclosed in December that Stanford University Medical School’s Jay Bhattacharya had been placed on a “trends blacklist,” which de-amplified his tweets during the pandemic. In other words, they used Web juju to limit the ability of others on Twitter to read what Bhattacharya had written.
It’s hard to measure precisely the cumulative effect of muting online dissent. But we know from the experience of dictatorships that governments that lack the feedback afforded by regular elections and free speech often forge ahead with reckless and stupid policies long past the point of futility. We also know that efforts to limit what the public can read, hear, or view are almost always done with the best intentions. The Soviets justified domestic censorship in part to protect their citizens from foreign disinformation. The nomenklatura of the Soviet Union believed they had an obligation to keep its radio, newspapers, and television free of Western propaganda.
None of this is to compare America today to the evil empire. Twitter and Facebook are private companies. And some content moderation is necessary for social-media platforms. But there is one important similarity between the original nomenklatura and our own. Like all censors, the content moderators today do not trust the rest of us to evaluate the information we encounter in the world. They believe that the minds of the masses are like balls of clay to be molded and shaped, incapable of critical thought and discernment. As history has proven time and time again, this assumption makes fools of us all.
We want to hear your thoughts about this article. Click here to send a letter to the editor.