My work aims to obtain independent evidence about algorithmic and AI systems and to expand the set of solutions available to policy makers. The audits my collaborators and I developed have shaped litigation, regulation, and platform practice in the United States and Europe, and our privacy research has changed how deployed systems protect users.
Equitable ad delivery in high-stakes contexts: housing, employment, and education. Our audits were the first to show that a platform's own ad delivery optimization, not only advertisers' targeting, skews who sees housing, employment, and education ads by race and gender (CSCW 2019, WWW 2021, FAccT 2024). The methodology supported the U.S. Department of Justice's first case challenging algorithmic bias, which ended in a 2022 settlement requiring Meta to change its ad delivery system; our 2025 external evaluation of the system Meta built in response (FAccT 2025 Best Paper) found that it reduces skew but at a cost to the advertisers that may be avoidable. Our work is cited in HUD's 2024 Fair Housing Act guidance on digital advertising, in Facebook's 2019 Civil Rights Audit, in a class-action charge before the Equal Employment Opportunity Commission, and in Equal Rights Center v. Meta.
Political advertising. In the United States, we showed that Facebook's delivery algorithms route political ads to users who already agree with them and charge campaigns more to reach everyone else (WSDM 2021), and that in the 2022 midterm elections advertisers used facially neutral interests as proxies for partisanship and race after platforms removed sensitive targeting options (CSCW 2024). In the European Union, the findings were presented in Piotr Sapieżyński's testimony before the European Parliament in 2022, and the European Commission's proposed regulation on the transparency and targeting of political advertising was amended in 2023 to cover ad delivery.
Auditing without sacrificing privacy. Platforms argue that external audits endanger user privacy. We showed that limited platform support, through differentially private outputs of a platform's own relevance estimators, lets auditors establish far more while protecting users (CSCW 2023). The result bears on how independent audits and researcher data access can be implemented under the EU Digital Services Act and other proposed legislation.
Changes in deployed systems. Our work has demonstrated that differential privacy is a feasible policy requirement and can be deployed at scale. On the accountability side, our demonstrations of privacy violations through microtargeted ads led to changes in Facebook's and Google's ad systems.
I have briefed the U.S. House Committee on Financial Services (2019) and the AI Working Group of the U.S. Equal Employment Opportunity Commission (2024), and I have spoken at the Federal Trade Commission.
In the United States, I co-signed CITP's comments to the National Telecommunications and Information Administration on AI accountability policy (June 2023). In March 2024, I joined researchers from MIT and Stanford in comments to the U.S. Copyright Office supporting an exemption that would let researchers circumvent account access controls to study bias and trustworthiness in AI systems; the Department of Justice cited those comments in its April 2024 letter to the Office.
In the European Union, I submitted feedback to the European Commission on the Digital Services Act's delegated regulation on independent audits (June 2023) and on its delegated regulation on data access for researchers (May 2023, including a joint submission with the Data Co-ops Project). In May 2026, I commented on the Commission's proposed measures for Alphabet's sharing of Google Search data under the Digital Markets Act.
I serve on the United Nations Independent International Scientific Panel on AI, which delivered its first preliminary report in July 2026, and on the Research Council of the International Association for Safe and Ethical AI.
Princeton's undergraduate minor in Computing, Society and Policy, launched in fall 2025, prepares students from any major to understand how computing systems shape society and to engage with the policy choices around them; I serve as its director.
My courses train the next generation of people who will build, regulate, and audit AI systems. SPI 585A, Societal Impacts of Data, Algorithms and AI, is a Master's seminar at the School of Public and International Affairs that pairs the technology behind data-rich systems with the policy and technical interventions that could change the status quo. COS 350, Ethics of Computing, is an undergraduate course that covers topics in algorithmic fairness, social media, AI safety, and privacy.
I have developed AI-auditing training and delivered it to several hundred U.S. government employees, and I have taught in the AI Training Series for Government Employees (2024) and the InnovateUS series on AI for public servants (2026).
