Internal documents show how Facebook polices speech, at times unfairly punishing and censoring black users or those in disputed territories like Palestine
In the wake of a terrorist attack in London earlier this month, a U.S. congressman wrote a Facebook post in which he called for the slaughter of “radicalized” Muslims.
Context & Ripple Effects
This 2017 ProPublica investigation is the earliest document in a leak arc that runs through the 2021 Facebook Papers coverage. It established two patterns that later reporting kept confirming: enforcement that falls hardest on Black users, and moderation of disputed territories like Palestine driven by under-resourced, English-centric rulebooks.
Four years on, internal documents showed the same machinery failing at scale — [[a:972152|non-English moderation gaps producing inflammatory language in Afghanistan and bans on common words in Palestine]] — while race-blind hate speech policies left minorities more exposed to racist language, not less. The through-line is that Facebook's global policy is written centrally but enforced unevenly.
First-order effects
- Black users and users in disputed territories like Palestine bear the direct cost: wrongful takedowns and censorship under rules applied inconsistently, even as a sitting U.S. congressman's call for violence against Muslims stays up.
- Facebook's own moderators are working from policies that its internal documents show produce these outcomes — meaning the company knew the asymmetry existed when it enforced against the affected users.
Second-order effects
- Each subsequent leak — the repressive-regime 'recognized crimes' list, the India dangerous-organizations failures, the Palestine and Afghanistan language gaps — compounds the reputational damage and hands civil-society critics and lawmakers a documented pattern rather than isolated anecdotes.
- Non-English-speaking markets become the visible weak point: where Facebook cannot resource linguistic and cultural context, its rules default to over-enforcement (banned common words) or under-enforcement (praise for mass killers), pushing affected governments to demand local control.
Third-order effects
- If the pattern holds, content moderation at platform scale structurally cannot be run as one centralized, race-blind, English-first rulebook — the endgame is either heavy investment in localized enforcement or external regulation and oversight bodies imposing accountability Facebook declines to self-apply.
- Moderation quality becomes a geopolitical variable: which speech gets censored in disputed territories and repressive regimes is decided by internal policy documents, making those documents matters of international contention rather than private corporate process.
The trend: Platform content governance is moving from opaque internal rulebooks toward forced transparency and localization, as leaked documents repeatedly show centralized moderation failing minority users and non-English markets.