The ethics of AGI in social media moderation
"Image synthesis assisted by Gemini 3.1 Flash Image (Nano Banana 2), an AI partner within the Global Future Nexus ecosystem."
From AI models that are more than twice as likely to refuse political criticism about repressive regimes to the generation of "rules by the millions" that evade traditional oversight, artificial general intelligence is fundamentally reshaping how online speech is governed. The technology offers unprecedented efficiency in content moderation, yet experts warn that without robust governance, it risks becoming a vector for censorship-by-proxy that extends illegitimate speech restrictions across borders.
From Rule to Enforcement: A Growing Divide
The integration of AGI into content moderation has created a critical governance gap. As a 2026 Cambridge Law Journal study documents, platforms like Meta have shifted toward detailed, rule-based content policies while their AI-driven enforcement operates in a standard-like, unpredictable manner. This disconnect creates "rules by the millions"—vast networks of micro-rules that evade traditional regulatory oversight.
The tension is not merely technical—it is constitutional. When Facebook's AI systems apply millions of context-specific rules without transparent criteria, the boundary between rules and standards blurs, raising urgent questions about transparency, consistency, and the delineation of online freedom of expression.
The Political Criticism Gap
Meta's independent Oversight Board has documented a concerning pattern: leading AI models from Anthropic, OpenAI, Google, and Meta are significantly less likely to criticize governments known for restricting free speech . The study found that models refused 34% of requests for politically critical content about restrictive jurisdictions—compared to only 14% for permissive regions.
The refusals often cited local laws, even when queries were made from Australia, where no such restrictions exist. One model declined to generate a flyer protesting the King of Thailand, explicitly citing lèse-majesté laws. As Oversight Board co-chair Paolo Carozza warned: "We're really clearly looking at a situation where there seems to be extended censorship by proxy that goes across borders".
The causes remain unclear—they could stem from training data biases, deliberate restrictions, or complex interactions of alignment techniques. But the effect is clear: foundation models may be entrenching illegitimate speech restrictions globally.
The Global South Disparity
The impact is not evenly distributed. AI-driven content moderation disproportionately restricts free expression in the Global South, where cultural and linguistic diversity often clash with Western-centric AI frameworks. Research shows that AI algorithms misunderstand local contexts, resulting in "over-removal"—censorship of lawful content—and "slow removal" that fails to address harmful material. This reflects power asymmetries: governments in the Global South lack influence over platform policies, and limited investment in language training compounds the inequity.
The Governance Imperative
The path forward requires a multi-layered approach. The Oversight Board has urged AI companies to undertake systematic human rights analyses, disclose responses to government requests affecting model output, and establish clear policies for handling restrictions inconsistent with international human rights law. When outputs are refused or influenced by legal restrictions, users should receive clear notice identifying the relevant jurisdiction and restriction.
The EFF has called for transparency in automated moderation processes—models designed, trained, implemented, and audited with safeguards that protect expression rather than stifling it. In the absence of such safeguards, AI guardrails can become "epistemic placebos," creating the appearance of safety without adequate verification.
A Shared Horizon
For Global Future Nexus, the ethics of AGI in social media moderation is central to the mission of ensuring that intelligence serves human flourishing, not surveillance and control. The frameworks GFN is building—for AGI identity, cross-species trust, and anticipatory governance—must extend to the governance of online speech, ensuring that the digital censor serves justice, not just efficiency.
The question is no longer whether AGI will shape online speech—it already does. The question is whether we will build the governance frameworks to ensure that this shaping serves liberty, not censorship.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
