Abstract:
This article, a chapter in a larger work on the global costs imposed by the actions and alignments of the period this book examines, addresses the cost to the information space, the shared record by which humanity understands itself. It documents a two-directional mechanism operating across the dominant platforms. In one direction, amplification: the rollback by Meta in January 2025 of fact-checking and hate-speech protections, permitting derogatory speech by national origin, against warnings from Amnesty International and Rohingya survivors that these were the conditions under which the United Nations found Facebook instrumental in a genocide against a Muslim minority; and the documented rise of hate on X after its moderation collapse, including posts amplifying anti-migrant violence to over one hundred million views. In the other direction, suppression: the report Meta itself commissioned finding Arabic content over-moderated while hate speech in Hebrew went under-moderated, the leaked designation list dominated by Arab and Muslim entities, and the documented systemic censorship of Palestinian voices. The chapter argues that a valve open to hatred of Muslims and closed to speech by them is not a malfunction but a design, and that its cost is a manufactured public perception of a fifth of humanity.
Main Article:
An honest information system can be wrong in any direction. Its errors scatter, favoring no one, because error without design is random. But when a system errs in one direction only, amplifying what harms a particular people while suppressing what they say in their own defense, the errors stop being errors. They become a valve, open one way and closed the other, and a valve is not an accident. It is a design. The dominant information platforms of the world have been documented, by their own commissioned reports and by the investigations of rights organizations, to operate as exactly such a valve with respect to Muslims, and this chapter documents both directions of its flow.
Begin with the direction of amplification, the opening of the gates to hatred, because it happened in public and on a date.
On January 7, 2025, the chief executive of Meta announced the dismantling of the moderation architecture of the largest information system in human history, ending third-party fact-checking, lifting prohibitions on speech previously banned as denigration of racialized minorities, rewriting the hateful conduct standard to permit exclusionary and derogatory language on the basis of national origin, and drastically reducing the automated moderation that had screened the flood of content. [1] The company framed this as the restoration of free expression. The organizations that had studied what its platforms had already done to one Muslim minority heard something else. Within days, a survivor of the Rohingya genocide, supported by Amnesty International and the Open Society Justice Initiative, filed a whistleblower complaint warning that the new policies recreated the precise conditions under which Facebook had fueled mass violence before. [2] For this is the documented history that hangs over everything. In Myanmar, the engagement algorithms of Facebook built an echo chamber of anti-Rohingya hatred, and a United Nations fact-finding commission found the platform had been instrumental in the radicalization of the population and the incitement of violence against the Rohingya, a Muslim minority, in a campaign the UN classified as genocide. [3] The gates being opened in 2025 were the same gates that had already, once, opened onto a genocide of Muslims, and the people opening them had been formally warned.
The second engine of amplification ran alongside the first, on the platform that had dismantled its moderation earliest.
After the purchase of X and the gutting of its trust and safety systems, the levels of hate speech on the platform rose considerably, by the assessment of legal scholars studying the moderation experiment. [4] And the amplification was not passive. The Center for Countering Digital Hate documented that three posts by the owner of the platform concerning anti-immigrant riots in Belfast generated more than one hundred and fifteen million views and were assessed as amplifying violent anti-migrant narratives that inspired calls to violence, riots which, as this publication documented elsewhere, targeted Muslim communities and were stoked by a funded network of anti-Muslim agitators. [5] The same owner distributed to his hundreds of millions of followers a film that the German regulatory board had refused to rate because it incited violence against migrants, a film whose vigilante hero hunts and kills targets the film identifies chiefly as migrants, and promised his followers a sequel. [6] The ideological meaning of that act belongs to another chapter of this work. What belongs to this one is the mechanics, that the largest private megaphone on earth was used to distribute, to an audience of hundreds of millions, a work judged by a state regulator to incite violence against the very populations, migrant and substantially Muslim, whose demonization the platform algorithms were already rewarding.
Now turn the valve over, and observe the other direction, the direction of suppression, because it is documented by the same companies own commissioned findings.
While the gates opened to hatred of Muslims, they closed against speech by them. The independent human rights assessment that Meta itself commissioned after the violence of 2021, conducted by the firm BSR, found that content in Arabic was over-moderated while hate speech in Hebrew went under-moderated, in part because the company had never built a Hebrew-language classifier to detect hostile speech at all, despite such speech being rampant on its platforms. [7] It found that the counterterrorism policy of the company, whose designation list, revealed in a leak, consists in its majority of individuals and entities from the Arab and Muslim world, disproportionately suppressed Palestinian speech, with penalties swifter and more severe than for any ordinary violation. [8] Human Rights Watch documented what it called systemic censorship of Palestine content, the removal of documentation of atrocities, the suspension of journalists, images of the wounded and the dead deleted under nudity policies, and concluded that the company had broken the very commitments it made after the previous round of documented bias. [9] Access Now documented that the detection threshold for inflammatory content was lowered specifically for content generated from Palestine, so that the algorithms flagged Palestinian speech on a hair trigger while Hebrew incitement flowed. [10] The direction of these findings never varies. The speech of the Muslim and the Palestinian is suppressed beyond the rules. The speech against them is permitted beyond the rules. That is the valve, documented from inside the company itself.
Hold the two directions together, because their combination is the finding of this chapter, and it is what neither direction shows alone.
A platform that merely failed at moderation would fail in all directions, harming everyone equally, favoring no one. What the documented record shows is the opposite, a consistent, years-long, structurally embedded asymmetry, in which the machinery loosens precisely where hatred of Muslims flows and tightens precisely where Muslims speak. Amplification one way. Suppression the other. The same company that could not build a Hebrew classifier to catch incitement against Palestinians lowered its detection threshold to catch Palestinians faster. The same ecosystem that reinstated the banned and unleashed the deniers throttled the journalists documenting a genocide as it happened, a suppression this publication has itself documented in the case of the artificial intelligence system suspended after naming that genocide. A one-way valve is not a malfunction. Malfunction is random. This is oriented, and orientation is design, whether the design lives in deliberate policy, in the incorporated bias of government terror lists, in the commercial cowardice that fears one lobby and not the other, or in the engagement algorithms that discovered hatred of a minority sells. The mechanism does not require a conspiracy. It only requires that every pressure on the system, political, commercial, and algorithmic, push the valve the same way, and the documented record shows that they do.
So name the cost, because it is larger than any single deleted post or amplified lie.
The information space is the instrument by which humanity perceives itself. Billions of people who will never meet a Muslim form their entire understanding of a fifth of the human race through what the machine shows them, and the machine has been documented to show them a systematically distorted picture, the hatred amplified, the reply suppressed, the caricature boosted, the human testimony deleted. This is the manufacture of perception at industrial scale, and its output is measurable in the world, in the riots stoked to a hundred million views, in the mosque attacked, in the genocide the UN traced to the feed. The Rohingya case is the controlled experiment history ran, the proof that the machine at full asymmetry ends in mass graves. And the cost falls on two parties at once. It falls on the Muslim, who must live inside a world whose image of him was engineered by a machine tuned against him. And it falls on everyone else, whose understanding of the world has been quietly corrupted, who believes his perception of a billion and a half people is his own when it was manufactured for him. A tax on the truth is paid by everyone who believes what he sees.
And here the boundary that governs this work applies, with the precision the documentation allows.
What is indicted here is a documented machinery and the documented policies that orient it, the moderation rollbacks announced on the record, the asymmetries established by the companies own commissioned reports, the amplifications measured by named research organizations. No people is indicted, and the chapter rests on the findings of institutions including the United Nations, Human Rights Watch, Amnesty International, and the assessment Meta itself paid for. Nor does the chapter claim a single controlling hand behind the valve. It claims what the record shows, that the pressures on the system all push one way, and that the result, whatever its mixture of policy, bias, cowardice, and profit, functions as a design. To document a machine that manufactures hatred of a people is the opposite of hatred. It is the defense of the people the machine is tuned against, and of the truth the machine corrupts for everyone.
So return to the valve, and read in its one-way flow the full measure of the cost.
An honest system errs in all directions. This one opens to hatred of Muslims and closes to their speech, by the finding of its own commissioned auditors, and it has already been found instrumental in one genocide of a Muslim people. The rules against hatred were loosened over the warnings of the survivors of that genocide. The megaphones amplified incitement to riot against Muslim communities to a hundred million views and distributed a film a state regulator judged to incite violence against migrants. And all the while, in the other direction, the testimony of the bombed and the besieged was throttled, deleted, and suspended, down to the artificial intelligence silenced for naming what it saw. The cost is a world that believes it knows what a Muslim is, having been taught by a machine tuned to teach it wrongly, and a shared human record corrupted at the source. That is the information tax, the one-way valve in the pipes through which humanity learns about itself, and it is paid every day, by the people it defames in the minds of billions, and by the billions whose minds are made up for them.
References:
1. Berkeley Technology Law Journal. "Meta's Fact-Checking Rollback: Governance, Free Speech, and User Safety," btlj.org, May 2025, on the January 2025 announcement ending third-party fact-checking in the United States; and Cambridge Forum on AI: Law and Governance, "Meta's AI Moderation and Free Speech," May 2025, on the revised Hateful Conduct standard permitting exclusionary and derogatory language including by national origin, the raised threshold for removal, and the reduction of automated moderation.
2. Amnesty International. "Meta's New Content Policies Risk Fueling More Mass Violence and Genocide," amnesty.org, February 2025, on the lifting of prohibitions on the denigration and harassment of racialized minorities, and the whistleblower complaint filed with the Securities and Exchange Commission on January 23, 2025 by the Rohingya survivor Maung Sawyeddollah with Amnesty International, the Open Society Justice Initiative, and Victim Advocates International.
3. Amnesty International, February 2025, and the cited United Nations Independent Fact-Finding Commission findings of 2018, on Facebook having been instrumental in the radicalization of local populations and the incitement of violence against the Rohingya, and the UN classification of the campaign as genocide.
4. Berkeley Technology Law Journal, May 2025, on the considerable rise in levels of hate speech on X after the dismantling of its content moderation team, the adoption of community notes, and the withdrawal from the European Code of Practice on Disinformation.
5. Center for Countering Digital Hate findings as reported in 2025 to 2026 coverage, on the three posts by the owner of X concerning the Belfast anti-immigrant riots generating more than one hundred and fifteen million views and amplifying violent anti-migrant narratives; and Fikr Institute prior documentation of the funded network behind the Belfast agitation.
6. Variety. "Elon Musk Posts Armie Hammer's Controversial Comeback Film Citizen Vigilante on X," variety.com, June 27, 2026, on the distribution of the film to the platform followers, the refusal of the German regulatory board to issue an age rating on the ground that the film incites violence against migrants, and the promise of a sequel; as analyzed in the companion Fikr Institute chapter on the ideological template.
7. Access Now. "How Meta Censors Palestinian Voices," accessnow.org, February 2024, and TechPolicy.Press, "A Perspective on Meta's Moderation of Palestinian Voices," May 2024, on the findings of the BSR human rights due diligence report commissioned by Meta, that Arabic content was over-moderated while Hebrew content including hate speech was under-moderated, and that the company had not developed a Hebrew-language classifier for hostile speech until 2023.
8. Access Now, February 2024, on the leaked Dangerous Individuals and Organizations list published by The Intercept, consisting in its majority of individuals and entities from the Arab and Muslim world, and the BSR finding that the policy disproportionately impacted Palestinian speech with swifter and steeper penalties.
9. Human Rights Watch. "Meta's Broken Promises: Systemic Censorship of Palestine Content on Instagram and Facebook," hrw.org, December 21, 2023, on the systemic censorship of Palestine content, the removal of documentation of atrocities, and the failure of the company to deliver on the commitments made after the 2021 findings.
10. Access Now, February 2024, on the reduction of the detection threshold for inflammatory content to twenty-five percent specifically for content generated from Palestine, and the continued reliance on tools the BSR report had shown to be discriminatory.