Blog

Over the past two centuries, white supremacists have created a sophisticated system of online organizing, utilizing the internet and its unique features to their advantage. The internet is an ideal breeding ground for white supremacy because it affords users a veil of anonymity, akin to the infamous “hood” Ku Klux Klan members wore to conceal their identity in the late 19^th and early 20^th centuries.[1] It is much harder to trace an individual’s involvement with a group online, especially for a lay community member.[2] This anonymity creates a level of privacy that allows individuals to explore fringe ideologies without fear of judgment and makes it easier to recruit new members.[3]

White supremacists also exploit the vastness of the internet and its ability to eliminate spatial boundaries to draw in socially, geographically, and politically isolated participants into a broader community.[4] White supremacists can recruit anyone who is interested, regardless of who they are, where they are located, or their level of familiarity with the ideology.[5] Then, they use their websites to connect these individuals from across the globe, creating a virtual white supremacist community from otherwise fringe pockets of extremists.[6] This creates the perception of “critical mass,” that members are part of a global movement much bigger than themselves, inspiring confidence and providing members with a sense of community.[7]

Though widely considered to be a fringe, extremist ideology, white supremacy has gained more presence and visibility, particularly in online spaces in recent years.[8] Seeing this, many people have been calling on social media platforms to implement stricter content moderation policies around hate speech.[9] However, despite efforts by social media platforms to exercise greater content oversight, content filters continue to let through extremist posts and remove benign and anti-extremist content. Though it is imperative that something be done around extremist organizing online, any technology, no matter how sophisticated, seems to backfire.

I. The Case for Content Moderation

Ideally, online content moderation serves two functions. First, it prevents uninformed users from being exposed to white supremacist content that could radicalize them to begin with. And second, it helps identify warning signs of white supremacist mobilization so these actions can be reported and shut down before they materialize into mass atrocities.

A. Preventing Exposure

Because of the reach of social media, we can now find and become part of groups with ideologies very different from those held by the people physically around us. Such exposure to radical content increases the likelihood that an individual will adopt the radical beliefs and the rate at which this radicalization will occur.[10] Similarly, and perhaps more potently, having online friends with radical beliefs also increases the likelihood that the individual will adopt a radical ideology.[11] Many white supremacist recruits found comfort in the bonds they formed with the individuals they met at rallies and online. The experienced members helped new recruits systematize and consolidate their beliefs into a coherent white supremacist worldview.[12]

Content moderation can be useful in ensuring that all such radicalizing content is scrubbed from accessible online spaces. That way, the likelihood that the average individual – one who isn’t searching for white supremacist content – would encounter these ideologies and become radicalized would decrease dramatically. 

B. Detecting Warning Signs of Mobilization

Although far-right extremist groups try to move in stealth, there are still several warning signs that indicate when they are poised to act that content moderation could help detect. Three such prominent warning signs are leakage, fixation, and identification.[13]

Leakage is when an individual communicates the intent to attack their target to a third party, commonly occurring in cases of targeted violence such as school shootings.[14] Fixation denotes a pathological obsession with a target or a cause.[15] Fixation can bolster an existing plan of attack or itself give rise to a plan as a byproduct of gathering immense amounts of information about the target.[16] And identification is when an individual aligns themselves with certain groups or values that justify or promote violent behavior. For example, individuals can identify with a “warrior mentality”, fighting for a “higher cause”, and seeing themselves as a “rescuer” for their community.[17]

II. Content Moderation in Practice

Content moderation has become a staple across social media platforms, ensuring that illegal, irrelevant, obscene, and insulting content is kept to a minimum.[18] However, although technological advances such as algorithms and filters have undoubtedly made content moderation more efficient, they have not necessarily improved its effectiveness. Automated content moderation tools still display significant blind spots that let large quantities of extremist content slip under the radar while penalizing non-violent and anti-extremist posts.

A. Gray Areas and White Supremacist Code

For all their purported effectiveness, content moderation protocols across social media platforms have failed to detect obvious warning signs of extremist mobilization that fall into gray areas. For example, Salvador Ramos, the gunman who slaughtered 19 children and two teachers at an elementary school in Uvalde, Texas, leaked a number of clues about his plans on his Instagram account.[19] Particularly, several days before the shooting, he posted a picture of a hand holding a gun magazine and (separately) photos of two AR-style semi-automatic rifles.[20] His TikTok profile also warned, “Kids be scared”[21]. None of these posts raised any alarm with the platform, however, because he didn’t make any direct, specific threats of violence.[22]

Because of the overabundance of content moderation tools, far-right extremists have developed a “code” to communicate with one another online.[23] This “code” relies heavily on dogwhistles, statements, and imagery that seem innocuous to an uninitiated observer but hold deeper, hateful meaning for members of a specific community.[24] For example, for white supremacists, the number 88 stands for Heil Hitler: H being the eighth letter of the alphabet and eight twice standing in for HH.[25] Some white supremacist circles refer to Muslim individuals as “skittles,” in reference to Donald Trump’s comment comparing Syrian refugees to “a bowl of poisoned skittles”.[26] This makes it incredibly difficult to regulate via automated content moderation because many of these numbers, images, and phrases are often used in harmless contexts completely unrelated to white supremacy. Because an algorithm cannot provide this necessary level of flexibility, white supremacist rhetoric will continue to fly under its radar, rendering social media platforms unable to completely weed it out.

B. Over-Policing Non-Violent and Anti-Extremist Content

On the flip side, while automated content filters let a large amount of extremist content through, their notorious rigidity frequently over-polices content that is not intended to be violent. They are especially poor at identifying a post’s tone (e.g., condemnation or sarcasm), cultural context, or non-violent radicalism. This heavy content moderation has also backfired by feeding into white supremacists’ narrative of victimization and causing many to migrate to less moderated platforms.

When it comes to tone, automated systems of content moderation are not flexible enough to account for the range of human emotion.[27] They are incapable of recognizing sarcasm or distinguishing whether a particular word or phrase is portrayed in a positive or negative light.[28] In this way, content moderation filters can (and frequently do) flag anti-extremist content for removal, despite the fact that these posts are calling out the hateful ideology, not supporting it. 

Automated content moderation is not always sensitive to cultural context, either. A Columbia University research team recruited Black and Latino youth who had experience with gang violence to identify whether specific Twitter posts were venting (expressing strong emotion), dissing (humiliating or degrading the audience), direct threats, callouts (inciting violence by calling the audience’s status into question), posturing, or rap lyrics.[29] Participants were split fairly evenly on their interpretations of each Tweet.[30] If youth with intimate knowledge of this community and the language they use can’t come to a consensus on the interpretation of potentially violent Tweets, how can one expect an algorithm operating from a rigid set of standards to do so?

Automated content moderation also does not differentiate between non-violent and violent radicals. Non-violent radicals support or justify extremist acts and may even express a willingness to engage in certain acts of violence.[31] Violent radicals, on the other hand, are the subgroup (less than 1%) of radicals that actually engage in extremist acts.[32] Because they hold the same ideology and use similar rhetoric when communicating these ideas, content filters that flag terms used by radicals generally will necessarily overrepresent non-violent radicals, who are statistically much less likely to act on their beliefs than violent radicals.[33]

And finally, in spite of all the good intentions behind content moderation on social media platforms, including rooting out and crushing extremist organizing, in many cases, these policies can backfire and strengthen these movements instead. These surveillance and apprehension efforts feed directly into white supremacists’ narrative of their victimhood and persecution, serving as a force to create camaraderie in the face of an “oppressor”.[34] The prevalence of content moderation on mainstream social media platforms has also prompted white supremacists to find and migrate to new platforms where they will be even harder to apprehend.[35]

III. Conclusion

If social media platforms continue to rely on automated content moderation, the white supremacy problem does not appear to have an effective technological solution. Therefore, the recourse in this situation may need to come from the legal realm instead. There has been significant discussion in the legal community around creating a private right of action for individuals harmed by white supremacist rhetoric online.[36] That way, hateful rhetoric has tangible financial and legal consequences for those who espouse it and those who harbor it, making it increasingly costly and risky to engage in.

One cannot wield a blunt tool and expect it to get sharper over time. Maybe it’s time, instead, to take out the sharpest tool in the tool shed and put an end, first to far-right extremism online and then far-right extremism once and for all.

---

* J.D. Candidate, 2026, Northwestern Pritzker School of Law. I’d like to thank Professor Meredith Rountree and my fellow staff members at Northwestern’s Journal of Law and Social Policy for their wise guidance and undying faith in the merits of this piece. I would also like to thank my sister, Osian. Our chronically online adventures sparked my interest in this topic and formed the basis of this blog post. And special shoutout to Learned Band for bringing me to law school on my hardest days.

[1] Kathleen M. Blee, Robert Futrell, & Pete Simi, Out of Hiding: Extremist White Supremacy and How It Can Be Stopped 40–41, (Caterina Froio, Andrea L. P. Pirro, & Stijn van Kessel eds., Routledge 2024).

[2] Corinne Segal, White Supremacists Once Wore Hoods. Now an Internet Mob Won’t Let Them Stay Anonymous, PBS News Hour (Aug. 20, 2017), https://www.pbs.org/newshour/nation/white-supremacists-wore-hoods-now-internet-mob-wont-let-stay-anonymous#:~:text=In%20the%20early%202000s%2C%20those,as%20an%20attempt%20at%20vigilante [https://perma.cc/9WP5-JJEJ].

[3] Blee, Futrell, & Simi, supra note 1, at 49.

[4] See Aaron Winter, Online Hate: From the Far-Right to the ‘Alt-Right’, and from the Margins to the Mainstream 39, 43 in Online Othering:Exploring Digital Violence and Discrimination on the Web (Karen Lumsden & Emily Harmden, eds., 2019).

[5] Adam Clark Estes, How Neo-Nazis Used the Internet to Instigate a Right-Wing Extremist Crisis, Vox (Feb. 2, 2021), https://www.vox.com/recode/22256387/facebook-telegram-qanon-proud-boys-alt-right-hate-groups [https://perma.cc/YPM6-UD97]; Winter, supra note 4, at 44–45.

[6] Winter, supra note 4, at 43, 47.

[7] Alexandra T. Evans & Heather J. Williams, How Extremism Operates Online: A Primer, Rand Corp. 1, 5 (Apr. 2022), https://www.rand.org/content/dam/rand/pubs/perspectives/PEA1400/PEA1458-2/RAND_PEA1458-2.pdf [https://perma.cc/L8GH-PRA5]; Benjamin Vaughan, Successful Online Communities Address the Primal Urge to Belong, Forbes Magazine (Oct. 21, 2019), https://www.forbes.com/sites/benjaminvaughan/2019/10/18/online-communities-and-the-primal-urge-to-belong/ [https://perma.cc/Q2KN-6HXZ] (discussing how humans are always seeking a sense of belonging and that by replicating this community feeling online, businesses can successfully draw customers and platform users in); Daniel Koehler, The Radical Online: Individual Radicalization Processes and the Role of the Internet, 1 Journal of Deradicalization  at 116, 121 (2014); Pete Simi & Robert Futrell, Cyberculture and the Endurance of White Power Activism, 34 Journal of Political and Military Sociology 115, 128 (2006).

[8] Amanda Seitz, White Supremacists Are Riling Up Thousands on Social Media, PBS News Hour (Jun. 10, 2022), https://www.pbs.org/newshour/politics/white-supremacists-are-riling-up-thousands-on-social-media [https://perma.cc/U8UH-BDA4].

[9] Cecilia Kang, It’s U.S. vs. World as Big Tech Faces Specter of Limiting Speech Online, N.Y. Times (Apr. 21, 2019), https://www.nytimes.com/2019/04/21/technology/facebook-zuckerberg-harmful-speech.html?searchResultPosition=17 [https://perma.cc/5FWH-GY87].

[10] Evans et al., supra note 7, at 12; Michael Wolfowicz, Simon Perry, Badi Hasisi, & David Weisburd, Faces of Radicalism: Differentiating Between Violent and Non-violent Radicals by Their Social Media Profiles, 116 Computers in Human Behavior at 1, 3 (2021).

[11] Wolfowicz et al., supra note 10, at 3.

[12] Blee, Futrell, & Simi, supra note 1, at 147.

[13] Katie Cohen, Fredrik Johansson, Lisa Kaati, & Jonas Clausen Mork, Detecting Linguistic Markers for Radical Violence in Social Media, 26 Terrorism and Political Violence 246, 248 (2014).

[14] Id.

[15] Id. at 249.

[16] Id.

[17] Id.

[18] See Andreas Veglis, Moderation Techniques for Social Media Content 137, 142, in Social Computing and Social Media (Gabriel Meiselwitz, ed., 2014).

[19] Amanda Seitz, Shooter Warning Signs Get Lost in Sea of Social Media Posts, AP News (May 27, 2022), https://apnews.com/article/uvalde-school-shooting-technology-shootings-social-media-texas-b8dc7a615765e17d46313bc83e2fe452#:~:text=WASHINGTON%20(AP)%20%E2%80%94%20The%20warning,19%20children%20and%20two%20teachers [https://perma.cc/6455-L4YL].

[20] Id.

[21] Id.

[22] Id.

[23] Gabriel Weimann & Ari Ben Am, Digital Dog Whistles: The New Online Language of Extremism, 2 International Journal of Security Studies at 1, 3 (2020).

[24] Id. at 8.

[25] Id. at 9.

[26] Id.

[27] Wolfowicz et al., supra note 10, at 2.

[28] Id.

[29] Desmond Upton Patton, Scott H. Decker, & William R. Frey, When Twitter Fingers Turn to Trigger Fingers: a Qualitative Study of Social Media-Related Gang Violence, 1 International Journal of Bullying Prevention 205, 208 (2019).

[30] Id. at 214.

[31] Wolfowicz et al., supra note 10, at 1.

[32] Id.

[33] Id.

[34] Winter, supra note 4, at 48–49.

[35] Blee, Futrell, & Simi, supra note 1, at 147; Matthew Grady & Rikar Hussein, White Supremacy Thriving Online, Despite Prevention Efforts, The Voice of America (Oct. 30, 2019), https://www.voanews.com/a/extremism-watch_white-supremacy-thriving-online-despite-prevention-efforts/6178570.html [https://perma.cc/2QDN-Z3CB]; Maura Conway, Ryan Scrivens, Logan Macnair, Right-Wing Extremists’ Persistent Online Presence: History and Contemporary Trends, International Centre for Counter-Terrorism – The Hague 1, 9 (Oct. 2019), https://icct.nl/sites/default/files/import/publication/Right-Wing-Extremists-Persistent-Online-Presence.pdf [https://perma.cc/AYV5-ZTRW].

[36] See generally, Engy Abdelkader, Elie Mystal, Wajahat Ali, Jim Weinstein, U.S. Elections 2020: Where and How Do We Draw A Constitutionally Permissible Line to A Candidate’s Inflammatory Political Rhetoric?, 44 Harbinger 140 (2020); See generally, T. Noble Foster & David W. Arnesen, Legal Strategies for Combating Online Terrorist Propaganda, 21 Atlantic L.J. 45 (2019) (these two articles are just examples of how the issue of hate speech and its legal limits is being discussed in the legal community).