Meta partners with YouGov and Oasis Labs to survey US Instagram users' race and make sure its products and AI operate fairly; Meta will only get aggregate data
AxiosIna Fried
Context & Ripple Effects
This partnership is the operational follow-through to Meta's late-2021 pledge, when it published a paper on how users from marginalized communities experience its products and signaled changes to how it uses US demographic data (that marginalized-communities review). The new step is concrete: YouGov runs the survey of US Instagram users' race, Oasis Labs holds the raw responses, and Meta receives only aggregate results — a structure designed to let Meta measure fairness without collecting the sensitive data itself.
The timing matters because the fairness mandate now explicitly covers AI, which is moving into Instagram's user-facing surface — the platform is already testing an AI-powered search bar that puts Meta AI in front of users. Aggregate-only third-party measurement gives Meta a way to audit those systems for disparate impact without owning the underlying race data.
First-order effects
Meta gets a fairness baseline for its products and AI systems across US Instagram users without holding individual race records, shifting sensitive-data custody to Oasis Labs and survey execution to YouGov.
Instagram users are the immediate affected party: some will receive race survey invitations, and their answers feed aggregate fairness metrics rather than individual profiles.
Second-order effects
The YouGov–Oasis Labs arrangement creates a template rivals can copy: platforms facing fairness scrutiny can commission third-party demographic measurement instead of building sensitive-data pipelines of their own, moving business toward survey and privacy-infrastructure vendors.
It also strengthens Meta's position in researcher-access debates — the company already runs a pilot granting researchers limited Instagram data access, and an aggregate-only fairness pipeline gives it a defensible answer to calls for more transparency without full data handovers.
Third-order effects
If the pattern holds, fairness auditing becomes an outsourced layer of platform operations — specialized intermediaries hold demographic data while platforms consume only aggregates, redrawing who is accountable when algorithmic systems produce disparate outcomes.
The same aggregate-measurement structure positions Meta to audit AI products at scale as they spread across its apps, making third-party demographic surveys a recurring cost of shipping AI to consumer surfaces rather than a one-off study.
The trend: Platforms are outsourcing demographic measurement to survey and privacy-infrastructure intermediaries so fairness and AI audits can run without the platform itself holding sensitive user data.
Race Measurement Survey 🗞 Starting today we'll ask a random assortment of people in the US to take an optional survey hosted by @yougovamerica where they can share their race/ethnicity. All responses are encrypted and can't be connected to your account. https://about.instagram.co…
also it's bull because when I asked @mosseri straight up about it .. he said something about different views and free speech. And everybody is doing surveys nobody is going back to honor the folks who caught hell for this @so_treu asked them straight up NOTHING https://twitter.co…
@UpFromTheCracks @hypervisible I'm *product manager hat on* why the new surveys and not a combined project with THE ACTUAL GOD DAMN USER REPORTS THEY HAVE
This voluntary survey is promising. We ask that @Meta engage with the civil rights community, particularly if any analysis will result in changes to products, policies, or practices. Given its troubling history on civil rights, we'll be watching closely. https://civilrights.org/.…
Oasis Labs Partners with @Meta to Assess Fairness for its AI Models This is a first-of-its kind initiative enabling inclusiveness and fairness Read more 👇🧵
I absolutely know there are amazing Latinx creators/ businesses - I follow them on TikTok. Would love for them to be highlighted more and more on Instagram. https://twitter.com/...
Really excited as the technology partner with @Meta on this first-of-its-kind initiative to assess fairness of its AI models for better inclusiveness, while preserving users' privacy, using cutting-edge privacy technologies, at such an unprecedented large scale real world usage! …
I know Meta basically has so much data on me already but I just don't trust giving them demographic information they ask for on their own out of “altruism” https://twitter.com/...
Stop what you're doing and start building on @OasisProtocol. A partnership of this magnitude is a huge vote of confidence for the unique privacy features of Oasis Network and it's sure to impact the lives of billions. https://twitter.com/...
More than a decade ago, I called social media the great equalizer, as no one need know your race or gender, and we could each be judged solely by the content of our character. How's that working out? https://twitter.com/...
Remember when Facebook asked people for their phone numbers for two-factor and then used the data to target ads? Anyway, think carefully about how much you trust this company's assurances before sharing any sensitive data. https://techcrunch.com/... https://twitter.com/...
Breaking my summer sabbatical from Twitter to highlight a launch that a bunch of teams at Meta including mine are very proud of: new privacy-protective surveys of IG users to enable us to conduct better fairness measurement around race https://about.instagram.com/ ...
This Meta AI blog post outlines how it all works on the back-end to ensure that we can do the fairness measurements we need without our ever having access to identifiable individuals' survey answers about their race: https://ai.facebook.com/...