When tech corporations announce layoffs, the accompanying linguistic framing is almost always sanitised: restructuring, optimisation, strategic reallocation, etc. But what disappears beneath these euphemisms is the human labour being deemed disposable. In May 2026, Meta announced slashing its workforce by 8,000, including employees from integrity teams, in favour of its AI ambitions. Later in June, reports emerged that the company is planning to replace 90% of its content moderation with AI. The move is not merely another corporate reshuffle in an industry intoxicated by automation, but it is also a warning about whose safety the tech giant is increasingly willing to negotiate away. Such decisions carry consequences that are felt most heavily by communities navigating political precarity, social exclusion and institutional neglect. At tech giants like Meta, integrity teams may be viewed as an expendable operational burden, but in the Global South, their absence and limitations can escalate into matters of public safety.

For years, safety and integrity teams across major social media platforms have projected a semblance of safeguard to digital public life, tasked with containing coordinated disinformation, hate speech, incitement to violence, and harassment before such material turns more combustible. There have been systemic failures, of course, including chronic understaffing, inconsistent enforcement, and slow or inadequate responses to appeals, especially in emerging or struggling markets that lack political economy and corporate control to initiate accountability. Integrity work has rarely been glamorous, and even less so equitable; moderation labour itself has long depended on underpaid workers, outsourced trauma and highly demanding tasks. However, the weakening of these systems at Meta presents a grim prospect for users in the Global South, given the company’s past moderation failures causing online abuse to morph into offline persecution with alarming speed on various occasions.

No learning from past mistakes

The dangers of hollowing out platform integrity mechanisms are neither speculative nor historically unprecedented. Following the dismantling of large portions of X’s trust and safety infrastructure after billionaire Elon Musk’s takeover in 2022, hate, misinformation and toxicity surged on the platform. Reinstatement of accounts previously suspended for violations of community guidelines, manufactured legitimacy through monetised verification, and Musk’s ideological repositioning of moderation itself as “censorship” contributed as catalysts. While these developments were largely framed around a twisted idea of “free speech”, the consequences were quite concerning.

The Center for Countering Digital Hate (CCDH) and Amnesty International, through protracted investigations, documented the rise in hate speech on X under Musk’s leadership. The CCDH found that the use of derogatory terms for gay men and transgender individuals had gone up 58% and 62% respectively. Additionally, a racist insult against Black people had appeared more than 26,000 times on the platform. X sued the CCDH in August 2023, claiming the non-profit had cost the platform “tens of millions of dollars” by causing leading advertisers to pull out. The lawsuit was dismissed in March 2024.

Similarly, a joint survey conducted by Amnesty International, GLAAD, and the Human Rights Campaign (HRC) documented an alarming spike in anti-LGBTQIA+ abuse after the platform’s trust and safety reductions. The investigation found that gender diverse activists and organisations experienced intensified harassment after Musk’s arrival, with a majority of respondents reporting an increase in hateful and abusive speech on the platform. Many also stated that reporting mechanisms yielded little to no action from the company. The investigations repeatedly challenged Musk’s claims that hate speech had declined under his leadership, documenting instead sharp increases in abusive and derogatory content targeting vulnerable communities. 

The X precedent ultimately revealed something far more consequential than the instability of a single social networking platform. It demonstrated how rapidly digital spaces can become hostile when corporations begin treating trust and safety systems as expendable obstacles to growth, profitability, or ideological experimentation. 

Meta’s moderation shortcomings across communities and borders

This brings us back to Meta’s influence in the Global South. Myanmar remains one of the clearest examples of how platform negligence can bring about real-world consequences. For years, Meta has been found complicit in anti-Rohingya propaganda in Myanmar. In 2017, extreme online hate triggered mass displacements, forcing more than 700,000 Rohingya to flee the country. Even the UN later acknowledged that Facebook had played a “determining role” in spreading hate against the Rohingya population. A 2018 Reuters investigation found at least 1,000 posts in Burmese targeting the Rohingya and other Muslims on Facebook in Myanmar. Six civil society organisations (CSOs), including Myanmar ICT for Development Organization (MIDO), wrote an open letter to founder Mark Zuckerberg, accusing Meta (then known as Facebook) of failing to invest adequately in moderation. Persistent coordinated campaigns against the Rohingya, inflammatory misinformation and genocidal rhetoric – much of which escaped meaningful moderation due to the company’s underinvestment in local-language expertise and contextual oversight – had exacerbated the situation. 

A similar pattern has emerged in India, where social networking platforms have repeatedly struggled to contain communal disinformation, caste abuse, and anti-Muslim incitement circulating on a vast scale. In 2024, Meta approved ads laced with derogatory language against Muslims and calling for their elimination. Investigation revealed that 22 ads had been submitted in Hindi, Gujarati, Kannada, Bengali and English, with 14 being approved for publication. Three more were greenlit following minor alterations, but those changes did not capture the intended incendiary messaging. Facebook has long been weaponised for propagating hate and fanning violence owing to moderation deficiencies in India. These accusations have not only come from civil society groups, but former Meta employees-turned-whistleblowers as well. Meta’s relations with the ruling Bharatiya Janata Party (BJP) have also made headlines for controversial reasons. A 2020 investigation by The Wall Street Journal (WSJ) revealed that Meta had been reluctant to enforce its hate speech rules against several BJP politicians, including one whose Facebook posts promoted fatal violence against Rohingya refugees and urged vandalisation of mosques. Meta was subsequently accused of deliberate inaction against content targeting Muslims on Facebook and Instagram to maintain friendly business ties with the government of India, which is its largest market in the world. In 2022, more than 20 CSOs, including APC, called on Meta to release its full Human Rights Impact Assessment (HRIA) for India amid concerns that the company’s platforms had amplified hate speech and incitement against minorities.

The implications for neighbouring countries such as Pakistan are equally alarming. Pakistan’s digital landscape already remains saturated with sectarian rhetoric, misogynistic abuse, blasphemy accusations and disinformation campaigns capable of escalating rapidly into offline harm. Journalists, women activists, religious minorities and transgender persons routinely confront targeted campaigns. Here, Meta’s Facebook and WhatsApp have a symbiotic link with X when it comes to hate and misinformation. Cases highlighting such incidents include persistent sexualised smear campaigns against journalist Asma Shirazi, who continues to face death and rape threats over her political views; politician Azma Bukhari’s deepfake in 2024 resulting in widespread hate, abuse, and misinformation; the far right-led disinformation campaigns in 2022 against the Transgender Persons (Protection of Rights) Act, 2018; and the 2014 violence against the Ahmadiyya community, which culminated in three deaths. Large volumes of derogatory Urdu-language hashtags and posts inciting violence continue to evade platform enforcement and frequently remain online despite user reports.

In Sri Lanka, consequences of platform oversight have exposed the illusion that social networking platforms just mirror social tensions rather than intensify them. In 2018, anti-Muslim propaganda, Sinhala Buddhist nationalism, and inflammatory misinformation proliferated unchecked across the platform. As communal violence escalated, the Sri Lankan government temporarily blocked Facebook, WhatsApp and Instagram, citing the spread of harmful content. But by then, digital hate had already bled into offline violence, deepening Islamophobia and social fragmentation, which reportedly left three people killed, 20 injured and 232 homes destroyed. A coalition of civil society groups, including the Women and Media Collective, expressed deep frustration with Meta’s inaction, arguing that local reports of inflammatory content had gone largely unaddressed. Two years later, Meta apologised for its role in the riots. According to Article One, the consultancy firm hired by Meta to investigate the link between its platforms and the violence, Facebook had “only two resource persons” to review content in the Sinhala language, which had been weaponised for mobilising coordinated attacks.

Meta’s moderation failures in Bangladesh have, time and again, illustrated how rapidly online misinformation can precipitate communal bloodshed. A stark example is the unrest that erupted during the 2021 Durga Puja celebrations. A Facebook post alleging the desecration of the Quran at a Hindu temple went viral, triggering a wave of violence against Hindu temples, homes and communities across several districts, leaving seven people killed and 150 injured. Bangladesh continues to witness outbreaks of violence linked to inflammatory Facebook content targeting religious minorities. Last year, a mob attacked the Hindu community over a Facebook post in Bangla by a teenager. Data from August 2024 to March 2025 show that 112 mob attacks led to 119 fatalities and 74 injuries, with many of these incidents either originating from content posted on Facebook or being amplified through the platform. 

Individual and collective safety suspended in corporate negotiation

The burden of weakened platform governance is never distributed equally. Those with institutional protection, political visibility and social privilege are often better positioned to withstand waves of disinformation and abuse. (According to the 2021 Facebook Files revelations, Meta earmarked 87% of its misinformation budget for English-language content, leaving other linguistically diverse markets with a meagre 13%.) Fringe communities, on the other hand, have to deal with these threats on their own. For them, platform deterioration can mean intensified surveillance, orchestrated intimidation, social exclusion and physical vulnerability.

Reports that Meta is planning to deploy AI to review and moderate vast volumes of content (although the company has neither publicly confirmed nor denied the reports) are concerning, as its failures to effectively mitigate harms on its platforms have already been widely documented. 

Moderation, in contexts discussed above, requires far more than mechanical detection of prohibited keywords. It demands cultural literacy, historical understanding and the ability to interpret how certain narratives acquire incendiary force within already polarised environments. Likewise, certain forms of gendered abuse and caste violence are sometimes embedded less in explicit vocabulary than in historical implication and social context.

The fantasy that AI can independently govern such spaces without substantial human oversight reveals not technological sophistication, but growing corporate impatience with the labour-intensive nature of meaningful moderation itself.

Moreover, shrinking of integrity infrastructure is driven by the economic demands of the contemporary AI race, where human moderation has been costly. Human moderation requires multilingual expertise, sustained staffing, psychological support systems and region-specific policy development. It also forces corporations to publicly acknowledge the immense volume of violence, exploitation and hate circulating across their platforms. AI, by contrast, offers Silicon Valley a seductive narrative of frictionless scalability: faster systems, reduced labour costs and automated governance capable of satisfying investors eager for technological expansion. Yet, the harms generated by weakened moderation do not disappear under automation. They are only redistributed downward onto those least equipped to absorb them.

As Meta deepens its AI ambitions, the company risks reproducing a familiar pattern within the platform economy: extracting immense value from Global South markets while treating the safety of those same users as negotiable. The deeper concern is what happens when corporations increasingly dismantle the human infrastructures capable of recognising danger before it mutates into turmoil. In politically volatile societies already struggling with disinformation, communal tensions and gendered violence, the erosion of the still insufficient safeguards does not herald technological progress. It forebodes abandonment masquerading as innovation.

Usman Shahid is a development professional and journalist working on digital rights, corporate accountability, trust and safety, and AI governance.