Platform rule: Facebook,corporate power,and networked publics

Introduction

In 2022, Amnesty International demanded that Facebook be held responsible for a genocide. Five years prior in Myanmar, the Muslim minority Rohingya

were killed, tortured, raped, and displaced in the thousands as part of the Myanmar security forces’ campaign of ethnic cleansing. In the months and years leading up to the atrocities, Facebook’s algorithms were intensifying a storm of hatred against the Rohingya which contributed to real-world violence. (2022)

The United Nations also implicated Facebook (Smith, 2018). Facebook itself agreed: “We know we need to do more to ensure we are a force for good in Myanmar, and in other countries facing their own crises” (Warofka, 2018). Facebook employees did not wield weapons against the Rohingya or burn their villages. No one from Facebook headquarters was even in the country. How can a platform in California be responsible for atrocities half way around the world?

Answering this question requires constructing a political history of Facebook that is attentive to how platforms exercise power over people at scale. Scholars typically study the links between societal harms and social media through research on behavioral micro-foundations, such as exposure to misinformation or polarization effects (Gauthier et al., 2026; Guess et al., 2023; Nyhan et al., 2023). Yet more sustained inquiry on the macro power of social platforms is also required that moves beyond individual behavior to understand the role of platforms in generating large-scale societal harms. A vast interdisciplinary literature critiques Facebook for mass surveillance (Zuboff, 2019), market consolidation (Galloway, 2017; Rahman, 2018), and public manipulation (Espíndola, 2025; Simons, 2023; Vaidhyanathan, 2018; van Dijck et al., 2019). Facebook executives echo these claims, warning, “God only knows what it’s doing to our children’s brains” (Field, 2017) and acknowledging that Facebook deploys “Big Tobacco tactics to make its products as addictive as possible” (Dodds, 2020). But platform impacts are difficult to parse across individual and societal levels: “Facebook likely has been—on balance—good for individuals. But Facebook has been—on balance—bad for all of us collectively” (Vaidhyanathan, 2018: 18). To understand this collective dimension, I trace how Facebook organizes and governs global collectives.

International Relations (IR) scholarship has highlighted Facebook’s role in global politics (Monsees et al., 2023) and algorithmic governance (Srivastava, 2023) as part of a broader focus on digital platforms as global political actors (Atal, 2021; Steinsson, 2024). But IR has yet to reckon with what social media scholars call “networked publics” that are “simultaneously (1) the space constructed through networked technologies and (2) the imagined collective that emerges as a result of the intersection of people, technology, and practice” (boyd, 2011: 39). Globally, it is hard to imagine a more influential source of networked publics than Facebook, the largest social platform with the longest history. Founded in 2004, Facebook’s over 3 billion users constitute more than a third of the world’s population. As the epigraph notes, no other entity, state or nonstate, has ever had direct access to this many people. Facebook is twice the size of TikTok and 10 times larger than Twitter/X. Facebook’s parent company, Meta, owns Instagram and WhatsApp (each with 2 billion users). While Meta’s most profitable market is the United States, where 68% of adults use Facebook and roughly half use Instagram (Schaeffer, 2024), recent growth has happened elsewhere. More than 77% of Facebook users reside in the global majority outside Northern America and Europe. In middle-income countries such as India, Brazil, Nigeria, South Africa, and Mexico, Facebook is the most used social platform (Poushter, 2024). In 65 countries, Facebook is the Internet through its connectivity program for mobile phones (Nothias, 2020). Facebook is also the largest media publisher and Meta leads the development of open-source Artificial Intelligence models. Thus, Facebook’s political history also offers guidance for how technology companies may exercise power in the AI era.

In this article, I conceptualize Facebook’s global power as platform rule. Platform rule refers to the transnational exercise of state-like rulemaking by digital platforms that structure the social terrain for how networked publics form and interact. Platforms as varied as eBay, Uber, and AirBnB make rules on “what is permitted and what is prohibited, who can interact with whom, what sorts of agreements are possible, and what kinds of rights and guarantees you have in practice if things go wrong” (Lehdonvirta, 2022: 3). Social platforms have direct access to their users, who make demands on how platforms are run (Gillespie, 2018; Klonick, 2018). Facebook makes rules on how to define harmful speech, whether to flag misinformation during elections, and how to handle government requests for user data, all of which have consequences for a polity of users. Facebook’s platform rule is global as decisions made in Palo Alto govern users in 180+ countries. Platforms often rule reluctantly (Barrett and Kreiss, 2019) and reactively to regulation (Gorwa, 2024). But the extent of rulemaking has turned social platforms into “special-purpose sovereigns who govern populations of end-users” (Balkin, 2018: 1193). Mark Zuckerberg also compares his platform to states: “In a lot of ways Facebook is more like a government than a traditional company. We have this large community of people, and more than other technology companies we’re really setting policies” (Srivastava, 2023: 990).

I identify three macro dynamics from theories of state power to understand Facebook’s platform rule. I first apply Mann’s notion of “infrastructural power,” where states exert power through controlling infrastructures that are useful for subjects. Facebook organizes “the ways people seek and share information, engage in debate, and participate in society” (Amnesty International, 2019: 11, Rahman, 2018; van Dijck et al., 2019). Zuckerberg (2017) pitches Facebook as infrastructure, saying “the most important thing we at Facebook can do is develop the social infrastructure to give people the power to build a global community that works for all of us.” IR too has turned to infrastructures (Bueger et al., 2023), including to understand emerging technologies (Bernards and Campbell-Verduyn, 2019). But Mann’s infrastructural power is about using infrastructures to create collective dependencies. From this perspective, Facebook is not neutral scaffolding for its users. Instead, Facebook wields discretionary power in how networked publics interact, an idea captured by Zuckerberg’s description of handling user privacy early on: “We decided these would be the social norms now, and we just went for it” (Levy, 2020: 267).

Second, I build on Scott’s concept of legibility through state modernization schemes to overview Facebook’s data surveillance and algorithmic manipulation. Facebook’s infrastructural schemes make users legible in ways that were not possible before by collecting ever more user (and nonuser) data and clustering individuals into networked groupings. Scott argues the purpose of legibility is for states to create governable subjects. For Facebook, the purpose of legibility is profit. Enabling new and different ways to make users legible for commercial purposes is how Facebook is worth $1.75 trillion while offering a free service. Extending that same legibility to political operatives is how Facebook tries to cultivate closer ties to governments that might otherwise be interested in stricter regulation. Facebook’s choice architectures also manufacture illegibility by not having appropriate tools to make users’ concerns seen or otherwise disadvantaging users from seeing what Facebook can see about them.

In the third dynamic, I apply Tilly’s conceptualization of the state as a racket that protects subjects from violence of its own making to Facebook’s relationship with users, especially in the Global South. Facebook did not set out to become a global platform. But social media’s growth imperatives incentivized Facebook to aggressively push into new markets, including those with large populations such as India and Brazil, and those without a saturated digital landscape, such as Myanmar (Wynn-Williams, 2025). Entering these markets generated problems for how the platform would apply its American-centric rules elsewhere. Facebook’s growth team followed the infamous motto adorned at company headquarters: “Move Fast and Break Things” (Levy, 2020). Users vulnerable to harms on Facebook come to rely on the “protection” offered by the platform’s rulemaking including content moderation. Yet there was “significant variation in content moderation resources afforded to different countries based on criteria that are not public or subject to external review” (Newton, 2021). Before 2021, only 13 percent of Facebook’s misinformation efforts were devoted to non-US countries (Cushing, 2021). Even though the US represents less than 9 percent of Facebook users, it took up 87 percent of protection resources. Thus, Facebook overextended operations in the global majority for its growth but did not invest adequate resources to mitigate the fallout.

Mann, Scott, and Tilly together help illustrate how Facebook’s power closely resembles the foundational society-shaping capacities historically associated with states. Each theorist addresses a distinct dimension of macro power relevant to platform rule. Mann theorizes how power operates through controlling essential infrastructures; Scott examines how power functions by rendering populations legible and thus manipulable; and Tilly explains how power sustains itself by generating and then “resolving” the problems that legitimize protection. These frameworks capture the full arc of platform rule from establishing dependencies, to extracting actionable data, to cultivating reliance on platform-provided remedies. Moreover, the macro power dynamics highlight that networked publics are not merely enabled by platforms but are actively structured, segmented, and differentially protected by platform rule itself.

Applying the macro dynamics of platform rule to Myanmar clarifies Facebook’s role in the atrocities. By 2016, Facebook had 18 million users in Myanmar—more than a third of the population—many accessing the Internet exclusively through Facebook’s connectivity initiative, Internet.org. The platform became infrastructural in a society where, as one observer noted, “people wouldn’t leave the shop unless the Facebook app had been downloaded onto their phones” (BBC, 2018). Yet as of 2014, Facebook employed only one Burmese-speaking content moderator worldwide (Wynn-Williams, 2025: 346). When senior leadership were informed of rising anti-Rohingya hate speech, they responded that investing in Myanmar was “not a priority” (Wynn-Williams, 2025: 350). At this time, Facebook was worth around $220 billion. The following sections detail how each macro dynamic operated in Myanmar and other contexts, often with life-and-death consequences.

In outlining platform rule, this article makes two contributions to IR scholarship on corporate power and private authority. First, it demonstrates that theories of state power illuminate the political capacities of digital platforms. IR scholars have long studied how corporations govern alongside or instead of states, emphasizing structural power in markets (Strange, 1996) and private authority in regulatory domains (Büthe and Mattli, 2011; Cutler et al., 1999; Hall and Biersteker, 2002). Yet these approaches tend to analyze corporate power as it operates within political systems—influencing rules, capturing regulators, or exploiting jurisdictional gaps—rather than as constituting the foundational conditions under which publics form and political life unfolds. Unlike most private governance, platforms are neither delegated governing authority by states (Avant, 2005; Srivastava, 2022), nor do they largely govern themselves in entrepreneurial self-regulation (Green, 2014). In platform rule, corporations govern publics of their own making. Thus, platforms such as Facebook should be treated differently from other instances of global market power that feature similar structural domination by firms but that do not involve governance of a polity. Rather than apprehend corporate power as state influence, my analysis foregrounds the state-like capacities through which platforms constitute and govern populations directly.

Second, this framework reveals how platform rule produces global hierarchies that systematically disadvantage the Global South—a dimension often obscured in accounts focused on platform power as consumer lock-in (Beaumier and Newman, 2025; Culpepper and Thelen, 2020) and regulatory arbitrage (Gorwa, 2024) in the Global North. Platform rule exhibits power asymmetries where firms direct resources toward wealthy markets while treating the global majority as expendable terrain for growth. A Facebook whistleblower tracking government fake accounts in the global majority observed:

There was so much violating behavior worldwide that it was left to my personal assessment of which cases to further investigate. . . . A manager mused that most of the world outside the West was effectively the Wild West with myself as the part-time dictator. (Silverman et al., 2020)

Using platform rule rather than governance to describe Facebook’s actions emphasizes the unevenness of how networked publics are treated. This dimension of platform rule resonates with IR scholarship on hierarchy, which has shown how international orders are structured by status distinctions, unequal authority relations, and differential protections (Hobson and Sharman, 2005; Lake, 2009; Zarakol, 2017). While this literature focuses primarily on states, platform rule reveals how private actors can also produce hierarchical ordering globally.

Ultimately, this article uses foundational theories of state power to situate Facebook as a paradigmatic case of platform rule. The unprecedented global scale of Facebook alone should attract more scholarly attention than it has received in IR, but the scope of Facebook’s rulemaking merits cracking open our toolkits typically reserved for macro histories of power. Facebook does not let academics into company headquarters for archival research. Yet its practices have left a public trail. I rely on three sources for the historical analysis. First, I systematically collected over 4,000 media reports on the company covering the past two decades. Investigative journalists have documented major Facebook scandals such as the Rohingya genocide. But media reporting is biased toward the Global North and early coverage was slow to understand dynamics of platform rule. Thus, I also use reporting from The Facebook Papers, a set of company documents leaked to journalists and the US Congress in 2021 by whistleblower Frances Haugen. Journalists say the documents show that “Facebook knows, in acute detail, that its platforms are riddled with flaws that cause harm, often in ways only the company fully understands. Moreover, the documents show, Facebook often lacks the will or the ability to address them” (Horowitz, 2021). (After the leak, Facebook rebranded as Meta.) Finally, I use extensive secondary literature in platform studies and employee memoirs (such as Wynn-Williams, 2025). The archive on Facebook is partial and only offers glimpses into platform rule. This partiality also makes Facebook more like states, which is precisely why historically versed IR scholars should study platform rule.

Infrastructural power

In economics, platforms represent two-sided markets connecting distinct user groups through an intermediary (Rochet and Tirole, 2003). Video game platforms are intermediaries between developers and users. Social platforms connect users to each other (the stated objective) and advertisers to users (the business model). Two-sided markets enable monopolistic gatekeeping through lock-in effects and self-preferencing (Beaumier and Newman, 2025). For instance, Amazon can force third-party sellers to make concessions to access prospective buyers on the platform. Separately, a large body of scholarship argues that platforms “have accrued considerable social and political value in addition to economic worth” (Caplan, 2023; DeCook et al., 2022; Helberger, 2020; Scharlach, 2024; van Dijck et al., 2019: 10). These scholars broaden the conceptualization of platforms from market intermediaries to sociopolitical infrastructure that provide “foundational goods and services on which the rest of society depend[s]” (Rahman, 2018: 1639). Facebook is infrastructure as it:

alters our ability to preserve and circulate ideas and stories, the ways in which we connect and converse, the people with whom we can interact, the things that we can see, and the structures of power that oversee the means of contact. (Marichal, 2012; Tufekci, 2017: 5; Vaidhyanathan, 2018)

Thinking in infrastructural terms recognizes that what makes Facebook significant is not simply its market dominance but that living without it “shackles social and cultural life” (Plantin and Punathambekar, 2019: 2). Moreover, the infrastructure view shifts “attention from markets as (ideally) level playing fields towards societal infrastructures, in which platforms introduce new hierarchies and dependencies” (van Dijck et al., 2019: 4). Unlike traditional media, “Facebook does not create content, it determines who sees what content. The power to design Facebook’s newsfeed is a more fundamental infrastructural power than the power to decide what to print or broadcast” (Simons, 2023: 138). The COVID-19 pandemic starkly revealed platforms’ infrastructural capacities. As governments struggled to coordinate responses, platforms became essential infrastructure for work, education, commerce, and social connection (Organisation for Economic Co-operation and Development (OECD), 2020). Facebook and WhatsApp served as primary channels for health information, while also becoming vectors for medical misinformation. In India, Twitter functioned as an emergency coordination system during the devastating 2021 surge, with users desperately seeking oxygen cylinders and hospital beds when state systems collapsed.

The dual identity of platforms as market intermediaries and productive infrastructures in “platform capitalism” (Srnicek, 2017) has roots in lean production and Toyotism, which restructured manufacturing through modular architectures, just-in-time logistics, and layered supplier networks (Steinberg, 2022). Toyota’s production system enabled both vertical coordination of suppliers and horizontal flexibility across markets. Platforms extend this logic by treating users themselves as both inputs and outputs: users generate the content that attract other users, while the platform orchestrates these interactions (Cusumano et al., 2019). Platforms can exercise infrastructural control across diverse sectors—from transportation to communication to commerce—by applying the same organizational logic wherever network effects and data extraction prove profitable.

Moving from infrastructure to infrastructural power is especially generative for studying platform dependencies. Michael Mann aimed to distinguish state elites from civil society power groupings, namely “ideological movements, economic classes, and military elites” (1984: 188). State elites exercise “despotic power” through a “range of actions which the elite is empowered to undertake without routine, institutionalised negotiation with civil society groups” (Mann, 1984: 188). In contrast, “infrastructural power” is “the capacity of the state to actually penetrate civil society, and to implement logistically political decisions throughout the realm” (Mann, 1984: 189). In despotic power, the state elites hold power over civil society groups, whereas infrastructural power is “collective power, ‘power through’ society, coordinating social life through state infrastructures” (Mann, 2012: 59, emphasis added). What does it mean to coordinate social life through state infrastructures? Mann writes:

The state can assess and tax our income and wealth at source, without our consent or that of our neighbours or kin (which states before about 1850 were never able to do), it stores and can recall immediately a massive amount of information about all of us, it can enforce its will within the day almost anywhere in its domains, its influence on the overall economy is enormous, it even directly provides the subsistence of most of us (in state employment, in pensions, in family allowances, etc.). The state penetrates everyday life more than did any historical state. Its infrastructural power has increased enormously. (1984: 189, emphasis original)

Mann points to the development of state infrastructures related to governance and administrative capacity (taxation, surveillance, justice, welfare, etc.) that affect more aspects of everyday life than before. Mann further uses the metaphor of a cage, arguing that “the ‘power’ of the modern state principally concerns not ‘state elites’ exercising power over society but a tightening state-society relation, caging social relations over the national rather than the local-regional or transnational terrain” (2012: 61).

Mann views the state as the only autonomous political actor able to centrally command people in a territory (1984: 208). As such, while nonstate actors such as corporations exhibiting economic power also develop infrastructures, Mann limits infrastructural power to state actors, insisting that unlike states we associate with corporations voluntarily. Yet the production of political power requires bargains among state and nonstate actors, for instance, in relations of “hybrid sovereignty” (Srivastava, 2022). Moreover, the voluntary nature of corporate–society relations is precisely what is contested when platforms become unavoidable infrastructure. As such, I argue that infrastructural power may be a source of political power for both state and nonstate actors.

The infrastructural power of platforms is cultivated through users. Amazon and Uber offer services that seem so efficient and convenient that their power “is exercised not against but often decidedly with a public that enjoys the fruits of innovation” (Culpepper and Thelen, 2020: 295, emphasis original). Social platforms in particular allow users to communicate in ways they could not before and with people they did not have access to with traditional communication as networked publics (Tufekci, 2017). In a basic sense, Facebook monetized its networked publics to become a “central bank for social capital,” giving the firm coercive leverage to “confiscate at will capital accumulated by users” by suspending accounts or downranking posts (Schwarz, 2019: 11). Confiscation of platform capital even applies to world leaders, as Trump learned after the 2021 Capitol insurrection.

But infrastructural power goes beyond confiscation to tighten user–platform relations. Media scholars use the idea of “affordances,” by which they mean what actions are afforded, or made possible, by platform choices. Social “platforms wield power by creating affordances and constraints through design choices and algorithmic curation that guide how people communicate and consume information, structuring social and political dynamics” (Scharlach, 2024: 2). In infrastructural power, while users are granted affordances, it is the platform that retains the power to grant. Uber’s affordances make it so drivers are not permitted to set their own prices or choose where they work whereas riders are afforded different rates and driver matches based on platform profiling (Lehdonvirta, 2022: 101). Facebook affordances are seen in design choices that limit user actions. A 2019 memo discusses disparities in user appeals to content takedowns: “Appeal rates are high in the U.S. + other similar countries (Germany, U.K., Canada) and very low in some other countries (India, Indonesia, Myanmar, and Cambodia). This is probably due to how our products are structured” (Gizmodo, 2022, “Appeal Rate by Country”). Facebook employees pointed out that Americans are four times more likely to appeal than Cambodians because of platform affordances. The German content moderation law, NetzDG, mandates that Facebook users can request removal of illegal content. But Facebook’s “NetzDG reporting form is hard to find and difficult to use, which helps explain its slow uptake by users in Germany” (Ahn et al., 2022: 2851).

Facebook’s infrastructural affordances herd users through AI recommender systems. Every second, Facebook curates information on Newsfeed out of thousands of potential posts. Whether users see a post from their family or are asked to friend a colleague or nudged to join a group, these social connections happen because of Facebook algorithms. In 2018, a Facebook employee noted the importance of algorithmically ranked content for the platform: “When we switch a random set of users to a pure chronological Newsfeed, their usage and engagement immediately drops” (Gizmodo, 2022, “Is Ranking Good?”). Such ranking can empower users-as-citizens to reach more people and bring down governments (Tufekci, 2017). Yet the same systems have been leveraged for information pollution (Smith, 2018). The 2018 employee memo lays it out:

Insofar as problematic content is often more engaging than unproblematic content, ranking-by-engagement runs the risk of favoring the problematic. . . . In this way, an unranked feed can provide a sort of herd immunity against misinformation. The misinformation might still spread, but that spread won’t be amplified, and may well die out before it gets too far. (Gizmodo, 2022, “Is Ranking Good?”)

Design choices around the 2016 US election illustrate how platform infrastructural power takes on political significance. In 2014, Facebook introduced Trending Topics, which amplified viral content on Newsfeed. Facebook had initially contracted journalists to oversee Trending Topics. But in spring 2016, after pressure from conservatives accusing Facebook of suppressing content, Facebook fired the journalists and turned over Trending Topics to algorithms entirely (Levy, 2020: 340–342). Trending Topics boosted visibility without accounting for reliability. This became evident in the circulation of fake emails from Hillary Clinton and the Democratic National Committee hacks by a “DCLeaks” account (Levy, 2020: 336). Three days before the presidential election, at least 140 pro-Trump fake news websites registered in Macedonia dominated Facebook (Silverman and Alexander, 2016). Trending Topics boosted these posts to half a million interactions, far exceeding engagement with The New York Times coverage of Trump’s taxes. The platform’s infrastructural affordances thus determined which information reached networked publics at a critical democratic moment.

Two days after Trump’s victory, Zuckerberg rejected that Facebook had any role in swaying votes through fake news, calling it a “pretty crazy idea. . . . Voters make decisions based on their lived experience” (Solon, 2016). Zuckerberg’s claim is at odds with his own desire to be a social infrastructure and indeed contradicts the lived reality of infrastructural power where platforms organize the conditions under which publics form and interact. Moreover, the company had internally called 2016 “the Facebook election” (Wynn-Williams, 2025: 256). In September 2017, Facebook confirmed reports that the Internet Research Agency, a Russian firm with links to the Kremlin, was behind 470 Facebook Pages and Profiles and 3000 Pro-Trump ads worth $100,000 (Frankel and Benner, 2018). Facebook’s recommender systems propelled the Pages, Profiles, and ads to reach more than 126 million Americans, 62,000 of whom pledged to attend “129 rallies and events meant to support Trump” (Vaidhyanathan, 2018: 88). While it is difficult to connect disinformation exposure to electoral outcomes, that Clinton lost by fewer than 47,000 votes in a swing state added to the perception of Facebook shaping the electorate. According to a former employee, Facebook’s senior leadership walked Zuckerberg through “all the ways that Facebook basically handed the election to Donald Trump” (Wynn-Williams, 2025: 264). I return to the 2016 election later, but here I emphasize how algorithmic curation tied users to platform infrastructures in ways that shaped democratic participation itself, a point echoed by Facebook internal documents: “With a ranked feed, Facebook decides what content spreads and what content doesn’t. . . . We are not a pass-through platform for users to share thoughts. We are aggregators, and this implies a very different set of responsibilities” (Gizmodo, 2022, “Is Ranking Good?”).

In Myanmar, only 1 percent of the population had access to the Internet in 2013, but nearly all of them were on Facebook (Wynn-Williams, 2025: 80). Facebook assumed infrastructural power by bringing Internet.org to a third of the population in partnership with telecom providers. Infrastructural affordances were different on Internet.org, a text-only version of the web that lacked “basic kinds of security and moderation,” meaning “terrorism, hate speech, fraud, spam, and sexual content all [went] unchecked” (Wynn-Williams, 2025: 202–203). Facebook had not posted Community Standards outlining its policies on hate speech in Burmese, the buttons for reporting violating content on the platform were not in Burmese, and Facebook’s infrastructure was incompatible with the Burmese typeface, meaning “for anyone outside Myanmar, the letters are just an unreadable jumble” (Wynn-Williams, 2025: 349).

Facebook further used platform affordances to both cozy up to politicians and confront them. A program known as Cross Check or XCheck shielded politicians from the platform’s normal enforcement where “some users are ‘whitelisted’—rendered immune from enforcement actions—while others are allowed to post rule-violating material pending Facebook employee reviews that often never come” (Horowitz, 2021). By 2020, XCheck had at least 5.8 million users who Facebook deemed “newsworthy,” “influential or popular,” or “PR risky” (Horowitz, 2021). XCheck allowed Trump’s summer 2020 post, “When the looting starts, the shooting starts,” to be left up on the platform even though it violated Facebook policy. A 2019 memo by employees, titled “The Political Whitelist Contradicts Facebook’s Core Stated Principles,” noted “We are knowingly exposing users to misinformation that we have the processes and resources to mitigate” (Horowitz, 2021). Moreover, XCheck gave an incumbency advantage as not all candidates running for office were whitelisted. Facebook has used infrastructural power against regulators. In 2015, India was deliberating whether to allow Facebook’s Internet.org, rebranded as Free Basics. All Indian users saw a pop-up that said, “Unless you take action now, India could lose free access to basic internet” (Wynn-Williams, 2025: 210). If a user clicked on the pop-up, it created a form letter to send to the regulator. Facebook also notified the user’s entire friend list that they sent a letter. People complained that “even if they declined to send the message, merely lingering on the page caused Facebook to send all their friends a notification saying they had written to the regulator” (Wynn-Williams, 2025: 211). Around 17 million user submissions were sent. India still banned Free Basics, a reminder that lobbying is not fully determinative of politics. Even when Facebook loses, though, we learn how its infrastructural power can harness user publics.

Legibility

The idea of legibility is advanced by James Scott for whom “high modernist” states are “devoted to rationalizing and standardizing what was a social hieroglyph into a legible and administratively more convenient format” (1998: 3). Scott argues that legible subjects are governable subjects, susceptible to “a high degree of schematic knowledge, control, and manipulation” (1998: 11). Scott ties legibility with manipulation:

Any substantial state intervention in society—to vaccinate a population, produce goods, mobilize labor, tax people and their property, conduct literacy campaigns, conscript soldiers, enforce sanitation standards, catch criminals, start universal schooling—requires the invention of units that are visible. The units in question might be citizens, villages, trees, fields, houses, or people grouped according to age. . . . Whatever the units being manipulated, they must be organized in a manner that permits them to be identified, observed, recorded, counted, aggregated, and monitored. . . .[O]ne might say that the greater the manipulation envisaged, the greater the legibility required to effect it. (1998: 183)

Scott says that societies exist before states in a diversity of illegible forms such as temporary encampments of hunter-gatherers, small peasant farms, communal property, customary water usage, and unregulated local naming (1998: 220). State ambitions to make permanent villages, large farms, centralized dams, and national identification require manipulating these pre-existing social groups, institutions, and practices through legibility practices. The greater the state’s ambitions, the more knowledgeable and intrusive it becomes (Scott, 1998: 184). Just like Mann, Scott focused on states. But he allows that “large-scale capitalism is just as much an agency of homogenization, uniformity, grids, and heroic simplification as the state is, with the difference being that, for capitalists, simplification must pay” (Scott, 1998: 8).

Facebook’s ambitions and legibility schemes are similarly intertwined in platform rule. It is in charge of managing 3 billion people to convene on the same platform and turn a profit. Facebook made networked publics legible to itself and others through simplification schemes used to turn people into data for microtargeting. Microtargeting allows narrowly tailored messaging based on real-world behavior (e.g. what a user has liked) rather than inferred preferences (e.g. where a user lives). For advertisers, “the more people engage with an ad, the less it costs” (Wynn-Williams, 2025: 265). The more legible someone is, the more profitable they are for Facebook and its clients. Data is thus central to the platform economy. Facebook’s business model is tied to data enhancing user legibility in two ways: (1) finding new data sources for users and nonusers and (2) creating new kinds of data by analyzing networked interactions. Platform data has “increasing returns to scale” meaning the more data a company processes, the more value it derives from each additional unit (Lehdonvirta, 2022: 120).

In pursuit of increasing returns to user legibility, Facebook extracted data from what users post and how they interact through clicks, Likes, Shares, Comments, hovers, deletions, and facial expressions. From the beginning, Zuckerberg argued for a pro-sharing social network, saying “You have one identity. . . . The days of you having a different image for your work friends or co-workers and for the other people you know are probably coming to an end pretty quickly” (Kirkpatrick, 2010: 36). In 2006, Facebook introduced Newsfeed. Previously, users visited friend profiles individually for updates, but Newsfeed gathered all updates and presented them to users in an endless scroll. Through Newsfeed, Facebook cultivated an “architecture of disclosure that encourage[d] users to share information about themselves” (Marichal, 2012: 7). In nascent stages of networked publics, “when others whom we consider close friends and family are posting intimate details about their lives, we are compelled to reciprocate or be excluded from the bonding process that is taking place” (Marichal, 2012: 54). These network effects are behind the longevity of social platforms even in face of user pushback.

Data extraction practices reveal how platform legibility extends far beyond the platform’s visible boundaries. Facebook did not restrict itself to what users voluntarily disclosed. It engaged in “surveillance capitalism,” where data is extracted in nonconsensual ways on what one does outside Facebook (Zuboff, 2019). In 2010, a researcher found that Facebook’s Like button was “installing cookies in users’ computers whether or not they click the button. . . . The button also tracks non-Facebook members, [concluding] that Facebook was potentially able to connect with, and therefore surveil, ‘all web users’” (Zuboff, 2019: 159). Facebook called the tracking a bug (Ghazali, 2011). By September 2011, this bug was available on one-third of the world’s thousand most-visited websites (Efrati, 2011). Later, a hacker discovered that Facebook tracked users even when they were logged off (Moses, 2011). Facebook called it a glitch, reflecting a wider pattern of dismissing surveillance mechanisms as technical errors that cultivates illegibility about its own practices while maximizing the legibility of users. As more than 10 million websites employ Facebook’s “Like” or “Share” buttons, the company’s tracking extends to even those without an account (Altaweel et al., 2015). Facebook used a feature where users tagged friends in photos to build one of the largest facial recognition databases in the world with the ability to identify obscured faces (Oreskovic, 2013). Facebook collects data from mobile phones, including apps with sensitive information on health or banking (Apple has restricted this tracking somewhat). A survey revealed that three-quarters of Facebook users are unaware of how much the company knows about them (Hitlin et al., 2019). This asymmetry—where platforms see users in ways users cannot see themselves or the platform—is central to how legibility functions as a source of power.

Facebook also extracted data from third-party apps, creating cascading legibility that extended far beyond individual consent. In 2007, Facebook introduced Platform, which allowed outside developers to create apps. With only 20 million users at the time, Facebook lured developers by promising them access to user data (Levy, 2020: 152–153). Platform was successful for Facebook’s growth. The popular game Farmville attracted 80 million players (Levy, 2020: 163). In 2008, a feature called Connect “allowed developers to use Facebook as a log-in on their own services and apps” (Levy, 2020: 169). Now Facebook was able to monitor user activities on thousands of third-party websites and applications. Moreover, Connect gave developers “Friends-of-Friends” access by providing the data of a user’s Friends who never signed up. Thus, Farmville could use Connect to access data for 160 Friends (average at the time) of each of their 80 million users. This architecture meant that legibility operated as a force multiplier: each consenting user rendered dozens of non-consenting users legible as well. In 2012, Facebook mandated that third-party websites and apps share user data with Facebook (Levy, 2020: 173). By 2014, Facebook wanted more control over the Newsfeed, which was being spammed by notifications from developers, and withdrew the Friends-of-Friends access. But it gave developers a 1-year grace period (Levy, 2020: 176). The same year, Facebook introduced Custom Audiences, which allowed advertisers to say, “Give me all the millennial women around Portland looking to buy a car” (Galloway, 2017: 106). Custom Audiences boosted Facebook’s legibility prowess to identify and microtarget individuals based on granular behavioral profiles assembled from data collected across multiple platforms and contexts.

Facebook’s choices for increasing legibility played a role in the Cambridge Analytica scandal. In 2014, researcher Aleksandr Kogan created a personality survey that recruited 200,000 paid respondents, but through the Friends-of-Friends feature Kogan ended up harvesting personal data from 87 million users (Levy, 2020: 409, 415). This data was passed to Cambridge Analytica, a political consultancy co-founded by Steve Bannon, which used Facebook’s Custom Audiences tool to build voter microtargeting profiles. Facebook prohibited developers from exporting user data but could not detect such misuse. The scandal illustrates how Facebook’s legibility schemes created vulnerabilities that political actors could exploit, which the platform had little capacity or incentive to prevent.

Facebook’s schemes further cultivate strategic illegibility in how users exercise rights in platform rule. Cambridge Analytica violated Facebook’s terms of service, which Facebook was made aware of in late 2015 from The Guardian. In January 2016, Facebook set up agreements with Cambridge Analytica to delete the data (Levy, 2020: 419). Cambridge Analytica did not certify it had done so until April 2017. Facebook did not require an audit to verify. Nor did Facebook inform the public or the 87 million users that their personal data was potentially misused in a consequential political election. Moreover, Facebook has tools that are obscure by design. Dark posts are “targeted nonpublic posts for specific users that only they would see” (Wynn-Williams, 2025: 209). Shadow bans limit the reach of posts so they may in effect be seen by no one or a limited number of users. Facebook’s legibility schemes are mobilized in an ecosystem “in which connections are invisible to the public eye and hence largely beyond societal control. The invisibility of centralised data flow control stands in sharp contrast to the user’s lack of control over their own generated data” (van Dijck et al., 2019: 8).

A related form of illegibility is that networked publics often fail to see themselves as such. Political theorist Jennifer Forestal argues that Facebook’s platform architectures are not designed

to facilitate our recognition that the content we see in these spaces is shared with others. And in obscuring our recognition of the interests we share, these invisible boundaries decrease the salience of the collective consequences of one’s activity on the platform. (Forestal, 2022: 155)

For example, Facebook shows users political content “not only because of their actual interest in politics, but also because their behaviors and the behaviors of their friends” leads to “an algorithmic interpretation of their interests as politically interested” (Thorson et al., 2019: 4). But “those ‘left behind’ cannot necessarily reassess and redress their previously expressed lack of political interest via new encounters with political content on Facebook” (Thorson et al., 2019: 12). The network behind networked publics has disproportionate information about what makes publics cohere. With sole access to its proprietary algorithms, publics are only legible to Facebook and clients of its choosing.

In Myanmar, Buddhist nationalists incited riots against the Rohingya in 2015. By then, Facebook had hired a second Burmese moderator, but he was suspected of “allowing a lot of racist content onto the site . . . [and] removing more posts by civil society groups and peace activists than government and anti-Muslim accounts” (Wynn-Williams, 2025: 351). Facebook’s security team was further aware of organized trolls hacking verified accounts and posting incendiary content, but the content operations team said the posts did not violate local laws and left them up (Wynn-Williams, 2025: 356). Violence escalated in August 2017 when military-aided attackers burned Muslim villages and carried out at least 6700 killings, systematic rapes, and drove over half a million Rohingyas into Bangladesh. Facebook had made the Rohingya legible as dehumanized targets to perpetrators through algorithmic amplification of hate speech. It made concerns of the Rohingya illegible by not hiring enough Burmese moderators and using a system that struggled with the typeface. In 2018, Facebook introduced automated translation of Burmese into English, but it was severely lacking. A Burmese post during the 2017 killings said, “Kill all the kalars [a slur for the Rohingya] that you see in Myanmar, none of them should be left alive,” which Facebook translated as “I shouldn’t have a rainbow in Myanmar” (Stecklow, 2018).

Legibility interacts with infrastructural power. Critics argue that Facebook’s engagement metrics are driven by emotional appeals that “offer frequent, low-level pleasures” (Marichal, 2012; Vaidhyanathan, 2018: 35; Zuboff, 2019). When microtargeting an electorate, Facebook algorithms may reward more extremist content. Social scientists have found mixed evidence for whether microtargeting creates extremist echo chambers (Bail, 2021). A study conducted before the 2020 US election found that majority of Facebook users see content from like-minded sources, but that reducing exposure to like-minded content had no measurable effect on affective polarization or ideological extremity (Nyhan et al., 2023). Related research discovered that chronologically ordered feeds (meaning, not algorithmically sorted) saw an increase in political and untrustworthy content on Facebook, but had no significant effects on affective polarization or political knowledge (Guess et al., 2023). Some challenge these findings from research on Twitter/X (Gauthier et al., 2026). Others cast doubt whether these studies, which partnered with Facebook and had employee co-authors, adequately accounted for Facebook’s limitations on untrustworthy content in the algorithmically sorted feed in the lead up to the election, distorting the control (Bagchi et al., 2024). Moreover, outside researchers do not have access to dark posts. In short, Facebook’s infrastructural power manufactures illegibility of its own practices for scrutiny.

Protection racket

Charles Tilly treated state-formation as a historical process that involves war making, fiscal extraction, and capital accumulation. As part of this process, Tilly likened states to organized crime. In standard understandings, a protection racket is where “a local strong man forces merchants to pay tribute in order to avoid damage—damage the strong man himself threatens to deliver” (Tilly, 1985: 170). In this way, a racketeer is “someone who creates a threat and then charges for its reduction” (Tilly, 1985: 171). Tilly then makes the classic argument: “To the extent that the threats against which a given government protects its citizens are imaginary or are consequences of its own activities, the government has organized a protection racket” (1985: 171). Just like war making and fiscal extraction gave rise to efficient militaries and tax administration, protection added “an apparatus by which the protected called forth the protection that was their due, notably through courts and representative assemblies” (Tilly, 1985: 181). There are two aspects of a racket. First, the racketeer protects people against problems of the racketeer’s own making. Second, the protected may make demands on the racketeer to meet the obligations of protection. Platform scholars have observed racket dynamics in a diffuse way. For instance, the crowdfunding platform GoFundMe presents itself as a “digital safety net,” while representing Silicon Valley “disruptors” that have “eroded our safety nets by promoting uninsured gig work and starving governments of tax revenue” (Lehdonvirta, 2022: 199).

Facebook’s protection racket is directly tied to its global scale and growth strategies. The transformation of Facebook into global public infrastructure was not inevitable. Facebook began as a website that assessed whether women were attractive, then became a directory for Harvard students and later the rest of the world. Pursuing a business model where the service is “free” (in pricing) required the platform to scale up quickly to capture more user data for its advertisers and grab market share from competitors. In 2012, as Facebook hit 1 billion users, senior management expressed concerns about “running out of road,” believing

the first billion users are the easy billion. . . . After that, you get into issues like how to reach children, how to reach parts of the world where there’s no internet, how to get into places like China that are hostile to any social media site like Facebook. (Wynn-Williams, 2025: 69)

Since then, as Facebook ballooned to 3 billion users, its platform allows the potential to “reach millions of people at once” and “easily connect many people who are not in the same physical space, or even people who do not know each other at all” (Tufekci, 2017: 6).

These newly networked publics generated problems for Facebook to manage (Klonick, 2018). Early on, moderation of inappropriate content on users’ Walls was not even considered (Levy, 2020: 112). It took Facebook 18 months to make “its first permanent hire for content moderation” (Gillespie, 2018: 118). Eventually, an internal document for “community standards” emerged in 2005 (Levy, 2020: 249). The standards expanded as Facebook grew. Apart from explicit images and violent content, Facebook was forced to reckon with online harassment, especially from “women and racial minorities, who argue[d] that the abuses have become so unbearable that platforms have an obligation to intervene” (Gillespie, 2018: 39). Social platforms today “develop systems for setting rules around what users can upload/share/comment/interact with, detecting content that might be breaking those rules, and ‘actioning’ content and enforcing potential violations by removing it, downranking it, or taking other actions” (Gorwa, 2024: 23, emphasis original). In content moderation, enforcement occurs through both overt content removal and account suspension and “visibility moderation” where platforms “manipulate the reach of user-generated content” (Chan et al., 2023: 1133).

The relationship between online harms and platforms is complicated. Hate speech was neither invented on Facebook nor carried out by its operatives directly. At the same time, while violence predates the state, being in the protection racket elevates the exposure level to violence. In a similar vein, Facebook’s internal research points to networked effects that have few offline comparisons: “hate speech on timelines generates more negative/hateful posts in response: when users producing [hate speech] are removed, the average hatefulness drops” (Gizmodo, 2022, “The Network Effects of Disrupting Hate Speech Enforcement”). Platform scholars argue that while “the relationship between platform, person, and society is a complex sociotechnical one, this relationship influences and is influenced by platform governance” (DeCook et al., 2022: 65). Facebook is then tasked with protecting users from hate speech and other online harms by accepting its “constitutive role in the emerging global public sphere” (Balkin, 2018: 1209).

Facebook’s platform rule thus began to look like protection of a polity of users. It had to “carefully construct an image of a responsiveness and attentiveness primarily concerned with responding to its user community, much like politicians must do with their constituents” (Marichal, 2012: 46). When describing rulemaking, platform “representatives often framed themselves in governmental or judicial terms” (Caplan, 2023: 3459). Facebook formalized rules in a plethora of user agreements where “the Terms of Service outline the ideal relationship between the user and the platform . . . Privacy Policies outline the ideal treatment of personal data. . . . [and] Community Guidelines outline the ideal experience of platforms as a community” (Scharlach et al., 2023: 6662–6663). In 2020, Facebook created an Oversight Board to hear user appeals, which was dubbed a “supreme court” by the media (Romm, 2019).

As networked publics appeared in new countries, polity problems compounded. What is considered hate speech in one jurisdiction may not apply to another. In early 2010s, “countries outside the U.S. became of vital importance” for Facebook’s growth (Wynn-Williams, 2025: 69). Facebook made a big push for India being a “testing ground” for emerging markets (Sharma, 2012). Facebook bet that if it became popular in India, it could expand successfully in Indonesia, Brazil, and countries in Africa. Indian regulators required platforms “to remove, within 36 hours of being notified, content that falls into a broad range of categories, including anything ‘grossly harmful,’ ‘menacing in nature’ or ‘ethnically objectionable’” (Sharma, 2012). At this point, Facebook had 65 million users in India, meaning the scale and scope of protection demanded from the platform were massive. (Today, with over 375 million users, India has the most Facebook users.) Facebook was similarly challenged by major protection demands in Indonesia, Myanmar, Philippines, Cambodia, and Sri Lanka, among others (these are the countries for which Facebook eventually released human rights impact assessments). To meet the demands, “internal documents show staff functioning like an internet age diplomatic corps, attempting to apply data science to the world’s thorniest conflicts” (Simonite, 2021).

But content moderation largely served as an area of control where Facebook could claim to do something about problems facing networked publics while limiting the scope of protection. Platforms exploit vagueness in moderation policies (Scharlach et al., 2023). No major platform, including Facebook, defines harm or harassment (DeCook et al., 2022: 67). Facebook, following YouTube and TikTok, justifies “policies primarily through references to physical and psychological impacts” (DeCook et al., 2022: 72). Platform protection fails when

narrow definitions of harm, violence, and danger are not just insufficient, but result in platforms engaging in ideological hegemony, imposing conceptions of not just what violence is and how it manifests, but who it impacts and by what mechanisms. . . . Rather than change the mechanisms of their design that enable harm, the platforms reconfigure intentionality and causality through policy documents in an attempt to stop users from being “‘harmful,” which, ironically, perpetuates harm. (DeCook et al., 2022: 65)

Platforms even avoid enshrining what constitutes as good. Public-centered values in platform rule include fairness, accountability, and transparency (Scharlach et al., 2023: 6661). Studies find that these “public and user-centric values were largely absent in platform policies” (Chan et al., 2023:1144, Scharlach et al., 2023).

Over time, platform rulemaking becomes unmoored from original problems. Facebook’s elections team, originally created in 2007 to help campaigns use the platform effectively, was restructured after 2016 to focus on election security. Facebook had embedded itself in political campaigns to make the electorate more legible: “Facebook’s tools and in-house white-glove service created incredibly accurate targeting of both message and audience, which is the holy grail of advertising” (Wynn-Williams, 2025: 265). After the Russian disinformation and Cambridge Analytica scandals broke, Facebook launched a “war room” monitoring threats to foreign microtargeting disinformation campaigns for major elections in the United States, Brazil, and France. By 2018, Facebook boasted to journalists that “each of its units—including Instagram and WhatsApp—has been told to make election security a top priority when designing products” (Frankel and Isaac, 2018). Infrastructural affordances then made the electorate legible for nefarious purposes, including massive data harvesting by Cambridge Analytica and Russian manipulation. To protect against these problems, Facebook created the war room and reinforced electoral integrity teams, but did not dial down its recommender systems, disentangle itself from campaigns, or lessen infrastructural control.

Facebook’s response to the 2021 US Capitol insurrection similarly sanitized its role. In 2019, Facebook announced it would allow false information in political campaign ads for the 2020 election (Isaac and Kang, 2020). Previously, it prohibited paid political ads that included debunked claims from third-party fact checkers. Facebook’s policy contrasted with Twitter, which banned all political advertising, and YouTube, which placed limitations on political ads (Isaac, 2019). In the run-up to November 2020, Facebook pledged to stop recommending all “political content or social issue Groups” that typically herd users toward extremist content, but researchers found this to be untrue (Yin and Ng, 2021). In the immediate aftermath of the election, Facebook recommender systems nudged users into numerous “Stop the Steal” groups, which its electoral integrity team were reluctant to shut down (Zakrzewski et al., 2023).

Facebook released its first Human Rights Report in 2022, conveying “significant investments in teams and technologies to better protect free and fair elections, including dedicated teams focused on election integrity and products that bring relevant and reliable voting information to people” (Meta, 2022: 7). Facebook did not say why the platform is taking actions on election security. Later, the company addressed the 2020 election, noting “we started preparations two years in advance. These enabled us to identify emerging threats and put systems in place to mitigate anticipated risks” (Meta, 2022: 48). There is no reference in the report to Russian interference or Cambridge Analytica. Instead, all platform rule is presented as context-free and indeed platform-free. A single bullet-point deals with the Capitol insurrection (i.e. Facebook worked with law enforcement to identify perpetrators). One watchdog group’s response alluded to racket dynamics, saying “the report fails to actually address the causes of the online abuses that pushed civil society to demand action from Meta in the first place. . . . The first step to solving a problem is admitting that there is one” (Marechal and Rydzak, 2022).

Facebook’s protection has developed in reactionary ways to external scrutiny. This leads to “platform transience,” the idea that “platform change is fast and continual, and as a result they are impermanent and ephemeral in significant ways” (Barrett and Kreiss, 2019: 2). After the 2024 US election, Facebook ended third-party fact-checking, relaxed restrictions around hate speech and harassment, and largely removed automatic flagging of violating content, arguing, “In recent years we’ve developed increasingly complex systems to manage content across our platforms, partly in response to societal and political pressure to moderate content. This approach has gone too far” (Meta, 2025). Advocates raised alarms that Facebook’s “new policy of minimal interference” is “about consciously choosing to be complicit in future atrocities whilst maintaining plausible deniability” (Hattotuwa, 2025). Even when there is less protection on platforms, it is the firms that control what to cut and how and what to disclose to the public.

Meanwhile, the Global South has been dealing with less protection from Facebook for years. Since Free Basics did not offer the same security or content moderation as regular Facebook, digital rights groups have accused Facebook of “delivering a crap version of the Internet to two-thirds of the world” (Wynn-Williams, 2025: 203). Arabic is the third most spoken language on Facebook. Yet the platform has consistently failed to invest resources for Arabic speakers. In 2020, an internal report said that “only 6 percent of Arabic-language hate content on Instagram was detected” compared to a 40 percent rate overall (Cushing, 2021). Another revealed that Facebook was incorrectly detecting terrorist content for Arabic speakers 77% of the time (Simonite, 2021). It is not just a matter of underpowered AI. A journalist who led the BBC’s Arabic News service “condemned Osama bin Laden, but Facebook’s algorithms misinterpreted the post as supporting the terrorist, which would have violated the platform’s rules. Human reviewers erroneously concurred with the automated decision” and the account was blocked (Horowitz, 2021). Arabic speakers were then both over-censored for dangerous speech and under-protected against hate speech.

Facebook’s protection extended to some countries more than others. In 2019, Facebook policy teams held a “Civic Summit” where they developed a tiered system for electoral integrity in which some countries were given more importance than others:

Brazil, India, and the U.S. were “tier zero,” the highest priority. For them, Facebook set up “war rooms” to monitor the network continuously, created dashboards to analyze network activity, and alerted local election officials to any problems. Germany, Indonesia, Iran, Israel, and Italy were placed in tier one. They would be given similar resources, minus some for enforcement of Facebook’s rules and for alerts outside the period directly around their elections. In tier two, 22 countries were added. They would have to go without the war rooms. . . . The rest of the world was placed into tier three. Facebook would review election-related material if it was escalated to them by content moderators. Otherwise, it would not intervene. (Newton, 2021)

For those deemed highest priority, the platform offered a full suite of AI classifiers, language support, dedicated staff, and a war room. The rest of the world might not have typefaces for outsiders to read or languages translated or fact-checking available. Platform rule does not affect everyone evenly; instead, “in the most vulnerable parts of the world—places with limited internet access, where smaller user numbers mean bad actors have undue influence—the trade-offs and mistakes that Facebook makes can have deadly consequences” (Cushing, 2021).

In Myanmar, Facebook embodied a protection racket in its growth-fueled decisions to expand without adequate safeguards and bad faith moderation that tied millions of vulnerable users to platform rule. Even though Facebook’s growth relied on users in the Global South, it neglected to take their lives seriously. As one employee told Amnesty: “Different countries are treated differently. If 1,000 people died in Myanmar tomorrow, it is less important than if 10 people in Britain die” (Amnesty International, 2022). Facebook’s Director of Public Policy described Myanmar as “a kind of lethal carelessness. At every turn, when Facebook’s leaders see how Facebook is inflaming tensions and making an unstable and frightening political situation much worse, they do . . . nothing” (Wynn-Williams, 2025: 346).

Four years after the Myanmar genocide, Facebook had not applied lessons learned in countries it deemed “at risk.” Facebook claimed that more than 90 percent of hate speech that it detects is taken down automatically, but that figure was only 0.2 percent in Afghanistan (Cushing, 2021). As the Taliban were taking over, Afghans had to individually flag hate speech and wait for review. Facebook also did not translate its reporting tools into Pashto and Dari. In Ethiopia, another site of civil conflict, Facebook’s classifiers fell short again. Employees bluntly assessed: “One of the primary challenges facing integrity work in at-risk countries is our ability to actually measure risks and harm. In many of these markets, we do not have any classification, and it can be exceedingly difficult to build” (Scott, 2021).

Myanmar, Ethiopia, and Afghanistan illustrate the costs when those furthest from power are last in line for protection. Hierarchy is an enduring feature of international politics (Zarakol, 2017) and this is not the first time that the Global South’s protection has been tied to proximity to power (Hobson and Sharman, 2005: 87)—the protector just happens to rule from board rooms.

Conclusion

This article constructed a political history of Facebook’s platform rule to highlight its infrastructural power, legibility schemes, and protection racket for billions of networked publics. It identified asymmetric power relations between Facebook and its users generally and how it treats the Global South specifically. The argument advances research on corporate power by foregrounding macro power dynamics as scholars have done in historical studies of state power.

What do these macro power dynamics reveal about platform rule? First, infrastructural power explains why platform dominance persists even amid user dissatisfaction: exit costs are prohibitive when platforms mediate essential social functions. Second, legibility highlights an asymmetry absent from market-focused accounts: platforms see users in ways users cannot see themselves or the platform. Third, the protection racket exposes a self-reinforcing dynamic, wherein platforms profit from the very harms they claim to mitigate. Taken together, these insights extend scholarship on private authority (Avant, 2005; Büthe and Mattli, 2011; Cutler et al., 1999; Green, 2014; Hall and Biersteker, 2002) and hybrid sovereignty (Srivastava, 2022) by specifying mechanisms through which platforms exercise not merely regulatory influence but state-like power over global populations. Globally, networked publics are not a unified public sphere but are fragmented and hierarchically governed, meaning hierarchical ordering in world politics (Hobson and Sharman, 2005; Lake, 2009; Zarakol, 2017) is not limited to relations among states. Platform rule produces its own hierarchies in which the Global South receive systematically inferior protections while bearing disproportionate risks.

Examining the world’s largest social platform contextualizes the sticky aspects of platform rule. Governments have mobilized unevenly against platforms (Bradford, 2023). Europe pursues a “rights-based” approach, but still relies on “privatized enforcement” where platforms play a pivotal role in implementing regulation. The United States engages in bipartisan tech scapegoating, but has yet to produce federal regulations for platform accountability. Platform rule may contain seeds for its own demise by following a cycle of offering less and less value for users (Doctorow, 2023). The largest platforms may shrink or collapse. Facebook’s growth in the United States has slowed in recent years relative to TikTok (but Facebook’s reach elsewhere remains impressive). Post-Musk Twitter has resulted in an exodus to BlueSky. From an infrastructural perspective, platforms should be more open to public critique as they assume central roles in society. But platform self-limitation requires that there are exit options (Beaumier and Newman, 2025) and that platform legitimation efforts are contestable (Srivastava, 2023).

In IR, the study of platforms has just begun. While this research focused on social media, scholars should also analyze Amazon, Uber, Spotify, and others to map their platform rule. Researchers should invest in longitudinal observation and cover greater breadth of experience by including non-American firms. A second strand could explore changing dynamics between platforms and their audiences that shape platform rule. This research could involve studies on platform–consumer alliances to test the responsiveness of platforms and specify how platforms cultivate different audiences. A final strand would extend platform rule to examine how AI development features the same firms as global political actors.