Last March, a community health activist in Myanmar published a warning about escalating violence. By Wednesday, it vanished, flagged, filtered, and disappeared by content algorithms she’ll never meet, applying standards she wasn’t shown, through processes no one can appeal. Halfway across the world that same day, a political influencer with ten million followers posted similar warnings. His remained visible, amplified, promoted.
These decisions join the 3.8 billion moderation verdicts algorithms make daily. According to industry data, 94% receive no human review whatsoever. More telling still: 28% of posts containing keywords like “protest” or “rally” face automatic temporary visibility restrictions, with the severity varying by geographic origin data.
Neither decision was explained. Neither was argued before judges or juries. There were no briefs filed, nor precedents cited, no dissenting opinions. Just lines of code executing their silent calculus of who deserves to be heard. We’ve replaced courthouse steps with server farms, legal arguments with machine learning parameters, and public deliberation with proprietary algorithms. This isn’t simply a technological upgrade. It’s a fundamental restructuring of how expression is governed, who holds the power to silence, and which voices carry. Content moderation algorithms don’t just enforce rules; they write the rules by deciding which enforcement patterns become normalized.
The EU Digital Services Act requires platforms to publish content moderation statistics. Fourth-quarter numbers showed 217 million removed posts globally. Researcher access to raw data remains restricted. Three platforms have faced preliminary fines for non-compliance with transparency requirements. Four have challenged these fines successfully.
The tech companies building these systems operate in a curious gray zone: too powerful to be mere platforms, too unaccountable to be proper governments, yet exercising authority that rivals both. Meta moderates more speech daily than all courts globally combined. Since 2021, Trust and Safety teams at major platforms have shrunk by an average of 72%, even as moderation decisions continue to multiply. This selective enforcement isn’t a bug in the system of algorithmic governance. It might be its defining feature.
Let’s take a closer look at who gets silenced and who doesn’t.
Twitter (I Refuse to Call It by That Single-Letter Rebrand) – A Broken Dream of Neutrality and Algorithmic Fairness
There was a lot of speculation around the intentions and goals of Elon Musk when he wanted to buy Twitter, which he rebranded as Twitter. It promised a “free speech absolutist” utopia. Instead, it exposed a stark reality: AI content moderation is a tool of selective enforcement. Musk swiftly reinstated banned accounts, including those of Donald Trump and conspiracy theorist Alex Jones, while touting algorithmic transparency. Ironically, the latter was even more controversial, considering Alex Jones’s involvement in the lawsuit regarding the Sandy Hook Elementary School shooting.
Yet, within months, Twitter quietly rolled back these policies under advertiser pressure, banning critics like journalist Aaron Rupar for “abusive behavior” while allowing far-right influencers like Andrew Tate to thrive. This naturally raised questions about Elon Musk’s commitment to free-speech, considering the double standards on its implementation on the platform.
This flip-flop epitomizes a broader pattern across tech platforms: public claims of neutrality mask opaque, politically convenient enforcement that privileges tech elites and MAGA-aligned voices. A separate yet still worth considering fact was that one of the first decisions after Twitter’s acquisition was firing roughly 80% of the staff (particularly engineers) working in Trust and Safety.
The worst part is this trend is not limited to Twitter. Other platforms are using content moderation to benefit certain groups while marginalizing others.
Meta’s New Content Policies
Meta’s content moderation has been a focal point of many human rights organizations, non-profits, and academics for sometime now. Its influence on human perception is enormous and we have seen several real-life examples. The platform’s algorithms are accused of fueling Myanmar conflicts and human rights abuses in 2017 (by Amnesty International) and the current policy changes are expected to fuel more turmoil.
However, there is another perspective on Meta’s new content policies. With all the focus on “Free expression,” Meta content moderation might follow the footsteps of Twitter, i.e., favoring the favorites. Another significant similarity to Twitter is that third-party fact-checking is being replaced with user fact-checking. One reason to take this grim perspective on Meta’s content policies is its storied history regarding content moderation:
Exception as Policy: Meta’s 2016 “newsworthy content” exemption allows rule-breaking posts if deemed publicly significant. In 2021, Trump’s posts inciting the January 6 riots were initially left up under this policy.
Inconsistent Application: While Trump’s account was suspended for two years, Meta permitted Brazilian President Bolsonaro to spread election fraud lies unchecked. Leaked documents reveal internal debates favoring “high-severity” politicians.
Whistleblower Revelations: Frances Haugen’s disclosures showed Instagram’s algorithms boosted anti-vax content from conservative influencers like Robert F. Kennedy Jr. while suppressing smaller accounts.
Considering this history and the new policies, it’s no wonder that the changes Mark Zuckerberg is implementing are being called a MAGA makeover.
Meta also tailors its AI moderation tools to respond to political and corporate pressure. During the 2020 US election, Facebook implemented strict anti-misinformation measures but rolled them back post-election, allowing the resurgence of previously flagged content. This demonstrates a pattern where enforcement fluctuates based on political calculations rather than consistent principles.
As unarguably the largest search engine in the world and the company behind the world’s largest video-sharing platform (YouTube), Google is uniquely positioned to shape perspectives and spread or deter misinformation. Unfortunately, its “hands” haven’t been clean as well. The AI moderation policies of Google and YouTube have also shown a pattern of selective enforcement.
YouTube’s demonetization loopholes: While small independent creators face harsh demonetization for policy violations, major media networks and influencers often get exceptions. High-profile content creators have been reinstated or given reduced penalties after public outcry despite violating the same rules that permanently banned smaller channels. It has also been observed on the two sides of the political spectrum, especially in 2017 when some leftist channels were demonetized (like Secular Talk) while other, equally aggressive views against climate sciences were allowed/left unchecked. It’s the AI era, but the attitude/approach is still concerning.
Selective Crackdowns: During COVID-19, YouTube banned alternative health accounts like David Icke but granted Fox News exceptions for pandemic misinformation.
Google has been more secretive about its algorithms compared to other platforms and it’s still taking a stance against EU’s fact checking requirements. In contrast, it has started implementing new DEI policies and scraping old programs since the new Trump Clown administration. The chances of this attitude seeping into content moderation are quite high.
OpenAI
OpenAI’s ChatGPT and other AI models have been criticized for inconsistent moderation. Initially, OpenAI enforced strict guidelines against misinformation and political bias, but pressure from tech elites and right-wing influencers has led to subtle changes.
Loosening restrictions: OpenAI has made adjustments to its moderation system following backlash from conservative influencers who claimed bias. This suggests that OpenAI, like other tech companies, is susceptible to external pressures that shape its enforcement policies.
AI-generated content moderation: While OpenAI claims neutrality, its models are trained on data that reflects biases inherent in the tech industry. This can lead to preferential moderation decisions that align with dominant narratives.
However, the biggest challenge related to an AI giant like OpenAI is not how it handles content moderation but the rise of leveraging AI models in content moderation on other platforms. AI has some inherent benefits in this regard, but if the creators and developers of these AI models show preferences similar to those of tech elites and politicians, the content moderation landscape might change for the worse. Despite the rift between OpenAI’s CEO Sam Altman and Elon Musk, Altman seems to have good ties with the White House’s current administration and is expected to benefit from the massive AI project “Stargate” worth about $500 billion (projected).
The Broligarch Network: Tech’s Shadow Governance
The term Broligarch was coined by journalist Kara Swisher. In general, it refers to a system of government where tech giants, or more accurately, the people behind them, have unprecedented access to and control over government policies. A simpler alternative term would be tech oligarchy. A more acute interpretation of the term would be a nexus of tech billionaires (Musk, Peter Thiel, Andreessen) and MAGA-aligned politicians (Josh Hawley, Ted Cruz). They can lobby for deregulation while demanding platform favors, essentially allowing them to wipe out or absorb the competition. One example would be DeepSeek, which has been growing at a pace comparable to ChatGPT. While it raises some legit privacy concerns and the governments are moving swiftly to ban it (at least on government devices), it has tilted the balance of open-source AI, which is worrying Broligarchs in this space, and they may even use their influence and make a move against it.
How Should AI Content Moderation Actually Be?
What we have discussed above are examples of selective content moderation, whether it’s purely AI or done with a degree of human intervention. The main concern behind these alarming content moderation decisions is the policy and approach behind how they are implemented. AI moderation can be a powerful tool to create a wide range of social media platforms that are not just safe for everyone but also an instrument of positive change. However, with the wrong approach, mindset, and self-interests, AI moderation can more than simply hamper and hurt free speech. It can silence and mitigate the voices of the opposition of the Broligarchs while magnifying perspectives and ideas they deem fit.
That said, it’s important to understand the potential of AI content moderation when implemented the right way. The most significant benefit is that once AI moderation is implemented following a guideline, it won’t misinterpret those guidelines as humans tend to do, ensuring consistent application of an AI content moderation policy.
Transparency & Accountability: Platforms should clearly define content policies, provide explanations for moderation decisions, offer appeals processes, and undergo independent audits to ensure fairness.
Minimize Bias & Ensure Fairness: AI models should be trained on diverse datasets, avoid over-reliance on automation by incorporating human oversight, and be regularly updated to adapt to evolving language and contexts.
Protect Free Speech While Curbing Harmful Content: Platforms should focus on removing clearly illegal or harmful content, differentiate between harmful misinformation and controversial opinions, and implement tiered interventions like fact-checking and labeling before outright removal.
User Control & Customization: Users should have the ability to customize moderation settings, adjust content filter strength, and participate in community-driven moderation tools for a more personalized experience.
Independent Oversight & Governance: Platforms should establish third-party review boards for high-impact cases, ensure independent oversight, and collaborate with governments without allowing political overreach or censorship.
It might seem too optimistic to hope for good and unbiased AI content moderation right away, especially when tech giants have such unprecedented access over regulatory bodies. But with awareness, competing technologies, and other governing bodies (like the EU) taking right steps in this regard, we may see a day when AI content moderation is used as an instrument of positive change, rather than selective enforcement.
Conclusion
We’re told AI will make systems fairer. Faster decisions, cleaner enforcement, fewer human flaws. But fairness without transparency is just control in disguise. What we’re watching isn’t progress, it’s a quiet transfer of power. Rules written in code, enforced in silence, and protected from scrutiny.
Content moderation was meant to keep things safe. It’s become a sorting mechanism. Promote the agreeable. Bury the rest. Policy blurred into preference. Platforms aren’t hosting speech—they’re shaping what exists.
And let’s not pretend the bias came from nowhere. These systems were built by people, trained on human judgments and calibrated to reflect the priorities of those in charge. The bias didn’t vanish—it was embedded, refined, and scaled.
The future of AI moderation won’t be fixed with better code. It depends on whether we’re ready to confront who’s pulling the levers. Because if we’re not, this won’t be innovation. It’ll just be obedience at scale.
Marc-Roger Gagné MAPP
@ottlegalrebels
To my number one fan—my father.
At 83, you’ve always lived in good health, steady and strong. And even now, fighting for your life against this sudden illness, you carry on with the same quiet strength. This is for you, with all the admiration of a son. Wishing you a full and speedy recovery. I still have more to write, and you’re still the first person I want reading it.
