ProPublica

ProPublica is a New York-based nonprofit investigative newsroom launched in January 2008 with a mission to expose abuses of power and betrayals of the public trust. Among U.S. investigative outlets, it produced the AI-harm investigations with the most durable policy outcomes: its Machine Bias series established algorithmic accountability as a journalistic discipline and drove a federal enforcement action against Meta's ad targeting system that reached the Department of Justice.

Founding and structure

Paul Steiger, former managing editor of The Wall Street Journal, launched ProPublica with funding from Herbert and Marion Sandler, who had sold Golden West Financial Corporation and committed $10 million per year to investigative journalism. The Sandler Foundation provided roughly 93% of ProPublica's funding in 2008, enabling a newsroom of twenty-four full-time reporters and editors at launch — the largest staff in U.S. journalism devoted solely to investigative reporting at that time. The organization has since diversified its funding base and grown to more than 150 journalists. It has won nine Pulitzer Prizes. Stephen Engelberg serves as editor-in-chief; Robin Sparkman as president. Herbert Sandler served as founding board chairman until 2016.

Machine Bias and COMPAS

ProPublica's "Machine Bias" series (launched 2015), led by reporters including Julia Angwin and Jeff Larson, established the template for empirical AI-bias investigation in journalism. The May 2016 investigation of COMPAS — a recidivism risk-scoring algorithm used in criminal sentencing across the United States — obtained more than 7,000 individual risk scores from Northpointe in Broward County, Florida, and documented that the system mislabeled Black defendants as future criminals at nearly twice the rate of white defendants, while systematically underestimating risk for white defendants. Northpointe contested the findings; the subsequent academic debate on algorithmic fairness — involving dozens of peer-reviewed papers — was itself a direct product of ProPublica's investigation, which made its full dataset and analysis code publicly available on GitHub. The Wisconsin Supreme Court, reviewing a sentencing case in which COMPAS had been used, ruled that risk scores cannot be determinative in sentencing decisions and that presentence investigation reports must include warnings about the accuracy limitations of such instruments.

Facebook housing ads discrimination

Between 2016 and 2022, ProPublica's Machine Bias team documented that Facebook's ad delivery algorithms allowed advertisers to exclude users by race, religion, national origin, disability, familial status, and other categories protected under the Fair Housing Act, Equal Credit Opportunity Act, and Age Discrimination in Employment Act. The October 2016 investigation found that housing advertisers could select ethnic-affinity segments as exclusion criteria; follow-up investigations found that Facebook had not eliminated the capability after its public promises, and that identical tools were enabling age and gender discrimination in employment advertising. In 2019, the U.S. Department of Housing and Urban Development filed a Fair Housing Act complaint against Facebook, citing the ProPublica investigations. In June 2022, the Department of Justice reached a settlement with Meta — the first federal case challenging algorithmic bias under the Fair Housing Act — requiring Meta to overhaul its ad targeting system and submit to court oversight. The settlement is the most consequential regulatory outcome traced to a single AI-harm investigation in U.S. journalism history.

Data publishing as movement infrastructure

ProPublica pioneered a model of investigative journalism that treats published datasets and analysis code as integral to the investigation — not supplementary. The COMPAS dataset release enabled years of independent academic and civil society scrutiny. Broader examples in the portfolio include the Surgeon Scorecard (2015), a public database enabling patients to query individual surgeons' complication rates from Medicare records, and Vital Signs, an API providing public access to physician payment and prescription-pattern data. This approach treats investigative journalism as a public infrastructure act: exposing the algorithm is step one; releasing the data enables an ongoing accountability ecosystem that outlasts the original story.

Relationships with the algorithmic accountability movement

ProPublica's Machine Bias investigations provided foundational empirical material for civil rights and algorithmic accountability organizations including the Algorithmic Justice League and Upturn. The Facebook housing ads reporting anchored HUD's complaint and the eventual DOJ civil rights action; the COMPAS investigation seeded the academic sub-field of algorithmic fairness, whose findings civil rights litigators and legislative advocates subsequently drew on. ProPublica has not formally partnered with civil society organizing campaigns in the way a grassroots coalition would, but its investigations have functioned as triggering events for movement action — a form of investigative journalism as movement infrastructure.