Sources and internal docs show Facebook's race-blind policies around hate speech left minorities more likely to see derogatory and racist language on the site

Last year, researchers at Facebook showed executives an example of the kind of hate speech circulating on the social network …

Washington Post 2021-11-22

Context & Ripple Effects

This report closes a loop that opened years ago. ProPublica's 2017 internal documents showed Facebook's speech rules were already punishing Black users disproportionately, and a follow-up study found the company itself conceded censors had erred on most posts submitted for review. The through-line is 'race-blind' moderation: rules written as if context-neutral, enforced at scale.

By late 2020, Facebook had begun reversing course, with leaked plans to police anti-Black hate speech more aggressively than anti-White comments. Today's reporting shows what the race-blind era cost in the meantime — Facebook's own researchers demonstrated to executives that minority users were seeing more derogatory language — and lands just weeks after internal documents showed the company's AI removes an estimated 3%-5% of hate speech.

First-order effects

Minority users on Facebook are the directly affected party: internal research shown to executives indicates they were more likely to encounter derogatory and racist language precisely because the policies treated all groups identically.
Facebook's policy and integrity teams face renewed internal and external pressure to complete the shift away from race-blind enforcement that began with the December 2020 leaked policy change.

Second-order effects

The finding collides with Facebook's enforcement model: its Q3 2020 transparency report showed 95% of hate speech takedowns were proactive, meaning automated systems catching only a small fraction of violations become the bottleneck for any identity-aware fix.
Regulators and civil-society critics gain concrete documentary evidence that neutrality-as-policy produced unequal outcomes, strengthening calls for audited enforcement rather than self-reported transparency metrics.

Third-order effects

If the pattern holds, large platforms move structurally from universal rulebooks toward group-aware moderation standards — a redefinition of platform fairness that will demand new transparency about how enforcement differs across communities.
The gap between what platforms claim to remove and what their AI actually catches becomes the central accountability question for content moderation, since policy changes without detection capacity leave outcomes unchanged.

The trend: Content moderation is shifting from context-neutral rulebooks to identity-aware enforcement, constrained by how little hate speech automated systems can actually detect.

Discussion

@lizzadwoskin Elizabeth Dwoskin on x
EXCLUSIVE: FB knew its hate speech algorithms disproportionately harmed Blacks and other minorities - but kept that info from its own civil rights auditor, new docs show. At one point 90% of autodeletions were comments attacking whites and men. https://www.washingtonpost.com/ ...
@washingtonpost @washingtonpost on x
Facebook's race-blind policies around hate speech came at the expense of Black users, new documents show https://www.washingtonpost.com/ ...
@rmac18 @rmac18 on x
This story has a lot, but the biggest takeaway has to be that Facebook lied to an independent investigator that sought to do a civil rights audit. I'm not sure anything surprises me about FB any more, but this shows the limits of self regulation. https://www.washingtonpost.com/ .…
@theldpage CUMmunist on x
Firstly - Black people* Secondly - if you wanted to be insta-banned/suspended from a Facebook platform all you had to post was “Men are trash”, the algorithm would almost instantly take you out. But terms like “faggot” and “moffie” are a free for all. https://twitter.com/...
@lorakolodny Lora Kolodny on x
Who sees their policy as “race blind”? Actively hiding and largely ignoring the data about, experiences and harms reported by Black people and others on FB- as the report says - sounds more racially biased than race blind to me. https://twitter.com/...
@paul_hundred Paul Hundred on x
But wait, it gets worse: it was racist for white people not to drop everything when black women complained Facebook wouldn't let them be racist against white people. https://twitter.com/...
@tarletong Tarleton Gillespie on x
So even if a conscientious group inside Facebook says “our site is loaded with racism,” real intervention is unlikely, because the solution - penalize attacks against minorities differently than against dominant groups - doesn't “seem right”...
@popehat ForbiddenPopehat on x
But muh Big Tech narrative https://twitter.com/...
@tarletong Tarleton Gillespie on x
What's more likely is exactly what Facebook did: train their algorithm to identify any attack on any group and penalize them all the same - regardless of its historical context, how much damage it can cause, or what other forms hate this overlooks or protects in the process.
@tarletong Tarleton Gillespie on x
But Facebook will continue to err on the side of “race-blind” policies until we win that bigger argument, outside of Facebook, beyond social media. Otherwise, it will continue to seem like the fairer approach, the neutral approach, the safer approach.
@tarletong Tarleton Gillespie on x
They will never design their tools to tip the scales of power - because they don't know any better, sometimes, but also because their metrics of success are conservative too, because they fear pushback, because when in doubt they look for solutions that “make sense”.
@jostevenslabour Jo Stevens on x
This is why the Tories' #onlinesafetybill embedding self-regulation by @facebook & others in law is the wrong approach. Self-regulation has failed over and over again. There is nothing in the Bill about algorithms which are at the heart of the problem we're trying to solve https:…
@slpng_giants @slpng_giants on x
No civil rights experts or @OversightBoard members worth their salt should stay with Facebook (@meta) after this. The company is using them as a meat shield to cover for their horrendous decisions. https://twitter.com/...
@wajahatali Wajahat Ali on x
Imagine if you were a Black employee of Facebook reading this news. What terrible and toxic leadership. https://twitter.com/...
@akelacooper Akela Cooper on x
All this horrible shit continues to come out about FB and yet the only thing that will dull its impact in the States is the fact that younger generations have other apps to occupy their time. https://twitter.com/...
@soledadobrien Soledad O'Brien on x
Hmmm. Someone should ask Sheryl Sandberg about this. https://twitter.com/...
@anylaurie16 Laurie Kilmartin- Austin Creek on x
What the fuck will make them stop? https://twitter.com/...
@mollyjongfast Molly Jong-Fast on x
Another story about how Facebook is a cesspool. https://www.washingtonpost.com/ ...
@qasimrashid @qasimrashid on x
Reminder: Facebook knowingly enabled the Rohingya genocide, admitted to it, & is now refusing to release data on the genocide So this anti-Black pro-racist approach doesn't surprise me one bit. I'm certain the full truth is much much worse than we can imagine. Break Up Facebook h…
@tarletong Tarleton Gillespie on x
We can criticize algorithms for being biased, and moderation policies for failing marginalized users. But this will never change until we can win the bigger societal argument: it's only “hate speech” when its punching downward. [thread] https://twitter.com/...
@drpanmd Dr. Richard Pan on x
“If you don't do something to check structural racism in your society, you're going to always end up amplifying it,” one of the people involved with the project told The Post. “And that is exactly what #Facebook 's algorithms did.” #EndRacism #StopHateForProfit https://twitter.co…
@zephyrteachout Zephyr Teachout on x
For starters: Break FB up from Instagram, WhatsApp, etc. (Horizontal breakup) and from messenger and other related apps (vertical). Monopoly power insulates FB from public/advertiser response to scandal. And ban the targeted ads business model where high conflict = profits.
@tarletong Tarleton Gillespie on x
This doesn't mean we don't criticize Silicon Valley. They've developed tools to innovate how society works, and all they've done is help us talk more and give our data away. They could do so much better. Especially when they're aware of the harm, as Facebook is. Reprehensible.
@normornstein Norman Ornstein on x
Obviously, heads should roll. Starting at the top. https://twitter.com/...
@pegobry Peg on x
So black people are more likely to express racial animus, and somehow this is taken as evidence that black people are victims? Am I reading this right? https://twitter.com/...
@bostonjoan @bostonjoan on x
But but but but... is just too much. This company does not understand what it has built and is run in such a way as to celebrate their own ignorance. It can't be reformed if they won't face the facts. https://twitter.com/...
@moonalice Roger McNamee on x
This story is soul crushing. I have spent five years criticizing FB policies without criticizing the people. This @washingtonpost story by the incredible @lizzadwoskin makes it clear that some FB executives are really bad people. https://twitter.com/...
@rexchapman Rex Chapman🏇🏼 on x
This is some seriously grotesque racism from Facebook leadership. https://twitter.com/...
@tarletong Tarleton Gillespie on x
I would love for Facebook to be ahead of the curve here, I want them to err on the side of civil rights. But I can't be surprised that, in a capitalist society, they will fail, because it reads as the riskier move. We have to make it the obvious move.
@naninizhoni @naninizhoni on x
In other words, “POC” produced a stunning 90% of hate speech on Facebook despite comprising less than 30% of American users. Ignore any charlatan or apologist who tries to rationalize this. The facts speak for themselves and they're unsurprising to anyone honest. https://twitter.…
@bridgetmarie @bridgetmarie on x
Appreciate the this reporting, but Black women have been saying this forever and it feels like it's only treated as “news” when white people say the same thing, only later. It's really frustrating. https://twitter.com/...
@jason_kint Jason Kint on x
This stat is absolutely horrific and new to me. We learned in prior reporting that % removed by automation was driven up by Facebook creating friction in the human review process but I didn't know it also was driven up by removing comments directed at white people. /6 https://twi…
@jason_kint Jason Kint on x
Here is Stone once again spreading misleading metrics which others regularly warn about onboarding and giving Facebook a platform to share them. % of content on the platform is irrelevant context for a report like this. /4 https://twitter.com/...
@jason_kint Jason Kint on x
The “worst of the worst” may better describe the governance, executive leadership and policy team that protects Facebook's toxic and harmful business model putting profits over democracy and civil societies. Andy Stone once again features in this report protecting Kaplan. /3 http…
@jason_kint Jason Kint on x
Even this sentence is deeply misleading. The word “proactive” is key as it means 80%+ ***of the removed hate speech*** is due to removal by software (as opposed to ~20% by humans). We learned from WSJ that maybe as low as 5% of hate speech is actually removed. /5 https://twitter.…
@jason_kint Jason Kint on x
This entire Washington Post report today is a must-read and once again deeply disturbing receipts on Facebook's willingness to allow toxic sludge in order to maintain global scalability. It sure seems like they lied during the most significant external audit in their history. /1 …
@jason_kint Jason Kint on x
Facebook even went to white people in the heart of deeply conservative America in order to establish content was objectionable. But even that apparently wasn't enough to get Bickert+Kaplan to support that hate speech against the most vulnerable groups needed to be prioritized. /2…
@jason_kint Jason Kint on x
It's the relentless accountability and investigative journalism by leading tech reporters like this which keep presenting me with new receipts at just how low Facebook will go for profits. https://twitter.com/...
@sifill_ldf Sherrilyn Ifill on x
We take no pleasure in having been proven right. We did our best to sound the alarm. At this point regulation is the only answer. https://www.washingtonpost.com/ ... https://twitter.com/...
@lizzadwoskin Elizabeth Dwoskin on x
In other words, FB did years of research into the racial bias of its algos while maintaining to civil rights auditors that it doesn't collect race data. They have their own side of the story here, and there is nuance. But this is the upshot to me. @nitashatiku @craigtimberg https…
@hypervisible @hypervisible on x
“Facebook largely operates with one set of standards for billions of users. Policies that could benefit a particular country or group were often dismissed because they were not ‘scalable’ around the globe” & might affect growth. https://www.washingtonpost.com/ ...

Chronicles