Algorithmic Justice League: Founder, Work and Funding

The Algorithmic Justice League is a Cambridge, Massachusetts group that audits commercial AI systems for bias, with face-recognition software at the center of its work. Wikipedia dates its founding to 2016, by computer scientist Joy Buolamwini. The group's own site doesn't give a founding date. It publishes no donor list, and no nonprofit registration turns up under its name.

Who Founded the Algorithmic Justice League and Why

Joy Buolamwini founded the Algorithmic Justice League while she was a graduate student at the MIT Media Lab, according to Wikipedia. The group credits its audits of commercial face-recognition accuracy with pushing some large vendors to pause police sales. Whether those audits caused the decisions is contested.

The group's own site confirms her role. Its about page lists her as "Founder and President" and says her experiences gave her "the motivation and insights to start the Algorithmic Justice League." The site doesn't give a founding date, so the 2016 year comes from Wikipedia.

The group is based in Cambridge, Massachusetts, and Wikipedia calls it a "digital advocacy non-profit organization." The about page does not describe it as housed at MIT. The about page mentions MIT only as the place Buolamwini did her graduate work.

The about page ties the group's start to her experiences. In 2018 she and Timnit Gebru published the Gender Shades paper, which tested three commercial gender classifiers. The group's library puts that paper at its center. You can read more about her work in our Joy Buolamwini profile.

What Algorithmic Justice Means to the Group

The about page doesn't give a one-line definition of algorithmic justice. It states a mission and four principles, and those show what the term means in the group's own work. The mission is to "raise public awareness about the impacts of AI" and to prevent AI harms.

The four principles are affirmative consent, meaningful transparency, continuous oversight and accountability, and actionable critique. The group also says it wants to equip advocates, build the voice of the most impacted communities, and engage researchers, policymakers and industry practitioners.

The method also shows the idea. An algorithmic audit buys or accesses a commercial system, tests it against a benchmark the maker did not pick, and publishes subgroup error rates the maker did not publish. The Gender Shades paper set that pattern. For background on the technology under audit, read our facial recognition guide. Our guide on whether AI is racist covers that question.

What the Algorithmic Justice League Does

The group does three kinds of work. It audits commercial AI products, it runs campaigns against specific uses, and it publishes research and reporting tools. Its library is organized into Research, Projects, Talks and Events, Policy and Advocacy, Exhibitions, Education and Press.

Audits and Research

The founding audit is Gender Shades. Buolamwini and Gebru built the Pilot Parliaments Benchmark, which includes 1,270 parliamentarians balanced by gender and Fitzpatrick skin type. They tested Microsoft's Cognitive Services Face API, IBM's Watson Visual Recognition API and Face++.

In 2019 Inioluwa Deborah Raji and Buolamwini published Actionable Auditing. It found that the August 2018 version of Amazon's Rekognition had a 31% error rate classifying the gender of darker-skinned women. Lighter-skinned men had a 0% error rate on the same test.

Testimony is another channel. Buolamwini submitted written testimony as Founder of the group to the U.S. House Committee on Oversight and Reform. The hearing, titled "Facial Recognition Technology: Part I," covered the technology's impact on civil rights and liberties.

Campaigns and Reporting Tools

The Safe Face Pledge launched in 2018 alongside Gender Shades. It was developed with the Georgetown Center on Privacy & Technology and urges organizations and governments to prohibit lethal use of facial recognition.

CRASH, launched in July 2020, proposes bug-bounty-style programs for reporting algorithmic bias. The project began in 2019 at a Rockefeller Foundation residency in Bellagio.

The 2021 audit of Olay's Skin Advisor, called "Decode the Bias," was run with ORCAA. It found bias by skin tone and age. Olay later took steps to audit and mitigate bias, and it funded 1,000 girls to attend a Black Girls Code camp. Nothing in the sources says whether the group was paid for that audit.

Other campaigns followed. DumpID.me, from January 2022, was a petition with Fight for the Future and EPIC against the IRS's use of ID.me. Freedom Flyers, from 2023, targets TSA facial recognition at U.S. airports, using a "TSA scorecard" and a report titled "Comply to Fly?" For the wider set of groups on these issues, see our guide to who is fighting AI.

The documentary Coded Bias premiered at the Sundance Film Festival and was distributed by Netflix in 2020. The about page says visitors can join through "Agents of Change," "Take Action" and "Share Your Story."

What Changed Because of Its Audits

The clearest documented outcome is the IRS's decision to halt its ID.me facial recognition program. The agency agreed in February 2022. The sources don't establish that the petition caused the decision.

The June 2020 retreat by vendors is the outcome the group claims credit for. IBM announced on June 9, 2020 that it would exit facial recognition, citing mass surveillance and racial profiling. Amazon announced on June 10 a one-year moratorium on police use of its facial recognition software. It said the pause "might give Congress enough time to implement appropriate rules."

Microsoft called for regulation but, as of June 11, 2020, hadn't said it would stop selling to police. Buolamwini's response was that "Microsoft also needs to take a stand." Those are the decisions the group points to when it claims credit.

The causal claim is where the record is weakest. Wikipedia links the 2020 pauses to the group's advocacy but says the link "does not establish causation." Contemporary reporting places Amazon's timing among protests over police brutality and the death of George Floyd. It says Amazon didn't cite the protests and doesn't connect Buolamwini's research to the decision.

The group also points to independent work. In December 2019 NIST published NISTIR 8280, which tested 189 algorithms from 99 developers. The study used 18.27 million images of 8.49 million people from State Department, DHS and FBI databases. In one-to-one matching, false positives for Asian and African American faces ran at "a factor of 10 to 100 times" the rate for Caucasian faces.

The NIST study is the independent check the group cites. The group's site says NIST "released a study that validated concerns" its research raised. Amazon claimed improvements afterward, but the sources found no independent verification of them. Our guide to company use of facial recognition covers the firms in more detail.

How the Algorithmic Justice League Is Funded

The group doesn't publish a full donor list. Its about page names no funders at all. It links only a donation page hosted on ActBlue, and it states no legal form, incorporation details or tax status.

Wikipedia names the Ford Foundation, the MacArthur Foundation, the Alfred P. Sloan Foundation, the Rockefeller Foundation, the Mozilla Foundation and individual private donors. It doesn't give amounts for any of them.

Tax filings name the group too. A ProPublica full-text search on October 21, 2026 returned 20 matching filings. They include Ford Foundation Form 990-PF filings for 2019 through 2023 and MacArthur Foundation filings for 2019, 2020, 2021 and 2023. Other matches come from the Patrick J. McGovern Foundation, the Heising-Simons Foundation, Democracy Fund, Warsh Mott Legacy and Code For Science And Society.

No dollar amounts were read in any of those filings. The group's revenue isn't published anywhere the sources found. A name in a funder's filing shows a connection, but it doesn't show how much money moved.

Its legal status is unresolved. ProPublica's Nonprofit Explorer returned "No nonprofits found" for the name on the same date. So no IRS-registered entity under that name turned up, and no Form 990 of its own. Wikipedia calls it a non-profit, but the sources don't establish whether it is a 501(c)(3), fiscally sponsored, or organized another way.

Nothing the sources found shows the group taking money from Microsoft, IBM, Amazon, Google, OpenAI or any other AI developer. That matters because the group has audited products from Microsoft, IBM and Amazon. Nothing found rules it out either, since no financial statement turned up to check. The Rockefeller link is a residency, not an AI company.

A reader who wants to give can use the donation link on the group's about page. The sources don't show how donations are spent or what the group's budget is.

Criticism of the Group and Its Research

Criticism comes from three places: the group's causal claims, the companies it audited, and scholars who question its framing. The group's own papers also admit limits. Each point below includes where to read the original.

The 2020 Retreats

The question is whether the group caused the 2020 pauses. Wikipedia links them to its advocacy, but it doesn't establish causation. Amazon's pause came while protests over police brutality were underway. Readers can compare the decisions and their limits in the Al Jazeera report from June 2020.

Vendor Pushback on the Rekognition Audit

Amazon disputed the 2019 audit. Its Matt Wood argued the study tested facial analysis rather than facial recognition, and that it used an outdated version of the service. He said the tests didn't match how the service is designed to be used. He pointed to an internal test at a 99% confidence threshold that produced zero false positive matches.

Other researchers defended the study in March 2019. A statement credited to "Concerned Researchers," posted on the MIT Media Lab site, said Amazon officials Matthew Wood and Michael Punke wrote blog posts that misrepresented the study. The group's site says more than 70 researchers defended its research. The group's reply on Medium states Amazon's objections in full, alongside its answer.

Limits the Authors Admit

The Gender Shades paper says its benchmark "is highly constrained since it is composed of official profile photos of parliamentarians." Subjects were cooperative, with fixed pose and lighting. The authors say results wouldn't transfer unchanged to real-world images, and they call for work on "an inclusive benchmark composed of unconstrained images."

The paper also says its categories are coarse. The vendors used the binary sex labels of female and male. The authors write that this view "does not adequately capture the complexities of gender or address transgender identities." They call the Fitzpatrick skin-type categories "fairly coarse" and defend them as "a scientifically based starting point."

Reform Versus Abolition

Os Keyes argues in "The Misgendering Machines" that automated gender recognition treats gender in a trans-exclusive way, whatever its accuracy. On that view, closing an accuracy gap doesn't remove the harm. Keyes has since argued for banning the technology rather than improving it.

Only the paper's title, author and argument were available, so the point stays general. The paper's entry on Semantic Scholar is the place to start. The wider ethics debate is in our facial recognition ethics guide.

Accountability and Self-Sourced Claims

The group demands "meaningful transparency" from the systems it audits. Its own site publishes no revenue figures and no funder list. No Form 990 of its own turned up either. A fair reading can name that gap without assuming bad faith.

Wikipedia notes that several Freedom Flyers and CRASH claims rely on the group's own materials. Its entry has no criticism section, so it isn't a balanced assessment. Readers who want the group's case in its own words can read its written testimony to the House committee.

Sources

• Algorithmic Justice League, About
• Wikipedia, Algorithmic Justice League
• Gender Shades paper, Proceedings of Machine Learning Research
• MIT Media Lab, on recent research auditing commercial facial analysis technology
• Buolamwini, response to Amazon Rekognition criticism
• Written testimony to the House Committee on Oversight and Reform, May 22, 2019
• NIST, study evaluates effects of race, age, sex on face recognition software
• Al Jazeera, Amazon pauses police use of its facial recognition software
• Semantic Scholar, The Misgendering Machines (Keyes)
• ProPublica Nonprofit Explorer, search for algorithmic justice league
• ProPublica Nonprofit Explorer, full-text search for the group's name