Oversight or Overreach? Free Speech and the Meta Oversight Board’s Latest Rulings
Recent decisions from Meta's Oversight Board show inconsistent application of international standards for free expression.
Why do some “offensive” posts remain online while others get removed? Meta’s Oversight Board, a quasi-Supreme Court for Meta platforms like Facebook and Instagram, has just handed down a batch of rulings that might raise more questions than answers. These decisions show how well-intentioned rules can punish the very voices they aim to protect and feed the societal tensions they hope to calm.
On April 23, the Oversight Board (OB) published its most recent decisions following Meta’s policy changes. Each ruling provides the public with insights into how some of the world’s biggest platforms enforce their speech policies. These decisions are always welcome, particularly when it comes to the broader status quo where social media platforms routinely remove ‘lawful but awful’ speech.
At the same time, the developments point to an inconsistency in treatment amongst different types of hate speech, raising significant concerns about equal protection and the broader implications for social cohesion. Using the principles of International Human Rights Law (IHRL), empirical data, and contemporary legal debates, this piece will examine the real-world consequences of these decisions and their implications for the future of free speech online.
Here’s the bottom line: establishing a high bar for limiting speech is an absolute necessity, not just a welcome recommendation. That high threshold needs to apply to all types of hate speech equally, be that racist speech or transphobic speech, for example. Without this framework, we risk stifling important public debate, discounting already underrepresented voices, and exacerbating an already polarized society.
The OB Sides with Free Speech in Cases Involving Gender Identity and the Apartheid Flag
In 2024, two Meta users misgendered transgender individuals in posts about bathroom use and female sports. These posts were reported for hate speech, but both were left up by Meta. Users appealed, prompting review by the OB, which supported Meta’s decision. Although the content involved misgendering transgender individuals, the majority of the OB found that, while the content in question could be considered offensive, it did not incite violence or discrimination, and the issue at stake was a matter of public concern.
In agreeing with Meta’s position on the case, the OB cited parts of the public comment put forth by The Future of Free Speech (FoFS), in which we urged the OB to adhere to Meta’s decision of non-removal. This position emanates from IHRL, specifically Article 20(2) of the International Covenant on Civil and Political Rights (ICCPR), which prohibits “any advocacy of national, racial or religious hatred that constitutes incitement to discrimination, hostility or violence'“ as well as the six-part threshold test set out by the Rabat Plan of Action (RPA).
In addition to law, we emphasized the importance of protecting open public debate on controversial issues, such as gender identity, warned against the democratic risks of censorship, and recommended consistency across Meta’s platforms. Suppressing public debate on issues like gender identity, even when it includes offensive language, risks chilling free expression on one of the most contested issues of our time. Significantly, this chill affects not only critics of prevailing norms but also transgender individuals themselves, who may feel less able to participate in discussion if the space becomes over-regulated or adversarial.
The OB decision referred to a case in May 2024, where a user posted a video on Instagram of themselves performing in a drag show, wearing a red, glittery outfit. The caption thanked collaborators and described themselves as a “faggy martyr.” Meta removed this post on hate speech grounds and reinstated it after the OB brought the case to Meta’s attention. This illustrative case highlights the challenges that marginalized groups face when it comes to content moderation. In response to stricter legislative demands and the threat of substantial fines, platforms often adopt a “better safe than sorry” stance, leading to overly stringent content moderation.
Furthermore, the use of AI may result in the biased enforcement of companies’ content policies. Whether it’s from a lack of data or biased training datasets, members of minority communities can potentially be silenced when AI platforms are unable to detect the nuance of language or linguistic reclamation. For the LGBTQ+ community, terms such as “dyke”, “fag” and “tranny” are a way of reclaiming power and a means for preparing members of this community to “cope with hostility.”
But this case shines a light on a deeper problem: the algorithms and policies built to spot “hate speech” often muzzle people who are simply speaking in their own voice. The trans community isn’t the only casualty. Tweets written in African American English are flagged as “offensive” almost twice as often as other tweets—clear evidence that bias still lurks in the code. If social platforms want to be the modern town square, they must guarantee that the very groups historically pushed to the margins can speak as freely as anyone else.
Similarly, in two cases involving images of South Africa’s apartheid-era flag, the OB upheld Meta’s decision to keep the posts online. While recognizing the flag’s painful legacy, the Board concluded that removing the content was not the least restrictive means available and affirmed that, even if offensive, the posts remained protected under international human rights law (IHRL). Interestingly, and reflective of the strong adherence to free speech that marked this round of cases, the OB’s majority decision to keep the posts up was made despite the finding that they breached Meta’s rules on hateful ideologies. It emphasized that this was done to remain in line with IHRL and called for greater clarity in how this Community Standard is defined and applied, reflecting a conflict between those rules and IHRL.
The OB’s Inconsistent Application When It Comes to Anti-Migrant Speech
In two cases involving anti-migrant posts from Poland and Germany, the OB overturned Meta’s original decision and ordered the content removed. The posts included a racist slur and sweeping claims about migrants as sexual predators. The majority concluded that, given the charged political climate ahead of key elections, the content posed a heightened risk of inciting discrimination and violence.
While the OB refers to the Rabat Plan of Action (RPA) and its six-part threshold test regarding the context of the post (in this case, the European elections), it did not conduct an analysis of the content in question under each of these points. While potentially unpleasant and offensive, it is unclear how this speech meets the threshold required for removal under Article 20(2) elucidated by the RPA.
Let’s take a look at each of the six RPA factors:
Statement Context: While election periods are often marked by polarizing immigration debate, and while this may raise sensitivity, not all heightened political tension justifies speech restriction.
Speaker’s Position or Status: While the status of the speakers (a political party and a public page) extends the message's reach, their institutional weight does not reach the level of state actors or influential figures, who could incite direct action.
Intent to Incite Audience against A Target Group: While inflammatory, the posts appear to constitute populist rhetoric rather than a direct call to action for discrimination or violence against particular groups.
Content and Form of the Statement: Although offensive and stereotypical, the content and form do not clearly advocate violence or hatred that satisfies the legal definition of incitement.
Extent of Dissemination: The extent of the speech act was not demonstrably wide, and there was no virality or mass dissemination associated with the content.
Likelihood of Harm: The lack of dissemination would lower the risk that harm would result from the speech.
While it’s beyond the scope of this essay to perform a complete analysis, it is clear that these factors were not fully considered in the OB’s judgment. Its decision noted the ‘unpredictable nature of online virality’ as a justification for Meta taking a more cautious approach to moderation. Notably, former Special Rapporteur on the Freedom of Opinion and Expression David Kaye highlighted that “international human rights standards can guide such policies, while the virality of hateful content in such contexts may require rapid reaction and early warning to protect fundamental rights.”
Nevertheless, this does not mean that the essence of the RPA must be diluted. More importantly, even if the OB believes these posts meet such tests, it would have been crucial for it to explain why. Pre-emptive speech restriction without adequate proof of imminent harm is not in line with IHRL, the standard by which the OB seeks to abide.
Elections, particularly in democratic states of the European Union, can by no means serve as a carte blanche for restricting offensive speech. In fact, Eric Heinze argues that in long-standing and prosperous democracies, characterized by sound institutions, “despite decades of pro-ban law and policy… no empirical evidence has, in any statistically standard way, traced hatred expressed within general public discourse to specifically harmful effects.”
The Problems with Restricting Hate Speech
Based on IHRL and other legal standards for determining the threshold of hate speech, the OB got two out of three decisions correct. While the gender identity and apartheid-flag cases reflected positive developments, the drop in speech thresholds when it came to anti-migrant speech does not show a consistent position by OB.
These cases reflect the legal, philosophical, and practical issues involved in regulating hate speech. Law Professor Eugene Volokh has coined the term “censorship envy,” referring to the resentment that arises when one group is shielded from criticism while others are not.
On a very basic level, the above cases could be interpreted as a protection of transphobic and white supremacist speech on the one hand but as a repression of anti-immigrant speech on the other. This inconsistent application can precipitate violence or social unrest, particularly when individuals perceive their expressive freedoms as being unduly restricted while other speech receives more protection. In such cases, some may resort to more confrontational or even violent forms of protest.
Within this context, protecting free speech, even when it is offensive or unsettling, can serve as a critical safety valve, potentially mitigating the risk of more dangerous outcomes. A key strategy in addressing far-right violence lies in reducing societal polarization, which implies that “openness and dialogue might work better than public repression, stigmatization or aggressive confrontation.” Research has shown that the rise of far-right extremism in Western Europe is rooted in a confluence of factors, including high levels of immigration, limited electoral success of radical right-wing parties, and the “extensive public repression of radical right actors and opinions.”
Moreover, restricting speech considered hateful, without meeting the strict tests of necessity and proportionality set out by IHRL, can lead to counterproductive results, such as pushing hateful attitudes underground. This allows extremists to adopt martyrdom narratives and contributes to the polarization of societies and social conflict through negative repercussions and backlash effects. While the OB’s citation of academic research was promising, its future decisions could benefit from the further use of such work, including empirical findings that reflect the actuality of situations.
Rather than embracing blunt instruments of censorship, content moderation should be based on IHRL and the RPA, provided that context does not, as was the case with anti-migrant speech, constitute a carte blanche for removal. Meta and the OB should, in detail, apply tests and thresholds to reach necessary, proportional, and, therefore, legitimate conclusions. Ensuring free speech on platforms in instances where the RPA conditions are not met protects the vital democratic function of speech since it allows for challenging, questioning, and dissenting opinions, even when they are offensive.
When hate speech poses (non-violent) challenges, especially for marginalized communities, censorship is not the answer. Censorship risks eroding democratic trust, stifling minority voices, and replacing open dialogue with enforced conformity. There is nothing stable, sustainable, or long-term in this approach. Quite the opposite. A free society cannot exist where controversial ideas are automatically equated with harm or, even worse, equated with harm in some types of hate speech (anti-migrant) and not others (transphobia and white supremacy).
In a free society, harm must be demonstrated, not assumed, and the default remedy for bad speech should be more speech, not silence. Counterspeech and counternarratives are essential tools for people to address harmful speech without censorship. Studies show that community-driven responses to hate speech, such as satire, rebuttals, or solidarity campaigns, are more effective and less polarizing than bans.
Above all, any restriction must be necessary and proportionate. As the former UN Special Rapporteur on Freedom of Expression emphasized, removing content should always be the least intrusive measure available, especially when dealing with controversial but lawful speech. Meta and the OB must take great care in how restrictions can backfire, leading to dismal consequences on individual, group, and societal levels.
Natalie Alkiviadou is a Senior Research Fellow at The Future of Free Speech. Her research interests lie in the freedom of expression, the far-right, hate speech, hate crime, and non-discrimination.