Hoover Institution (Stanford, CA)— Hoover’s Technology Policy Accelerator (TPA) and Stanford’s Institute for Human-Centered Artificial Intelligence are funding three new research projects that explore how the use of AI can both illuminate and complicate ongoing geopolitical challenges.
As part of Stanford’s seed grants portfolio, HAI and TPA jointly awarded funding of up to $100,000 each to three new research projects that explore nuclear proliferation, public perceptions of AI, and the political differences identifiable in AI agents impact our understanding of AI, and the effect its rise will have on matters of geopolitical significance.
Each proposal comes from an interdisciplinary team including at least one principal investigator from the humanities or social sciences with at least one principal investigator from a technical field.
“AI is rapidly reshaping the geopolitical landscape, but policymakers still lack the research needed to understand where the greatest risks and opportunities lie,” said Amy Zegart, the Morris Arnold and Nona Jean Cox Senior Fellow at the Hoover Institution and TPA’s director. “These projects bring technical and policy experts together to examine urgent questions about nuclear security, public opinion, and the use of AI in political decision-making. By sparking new collaborations across fields, this partnership will generate rigorous research that can inform policy before the technology moves even further ahead.”
The partnership between both groups began in early 2026, when Stanford’s HAI and Hoover’s TPA formed a new Working Group on AI and Geopolitics to advance an ambitious research agenda focused on issues at the intersection of these two concerns.
The working group’s goal is to develop orienting principles and fund novel research about the most important geopolitical policy challenges impacted by AI. Its participants focus on understanding the potential national and international security, economic, and policy impact of technical breakthroughs that should be anticipated in the coming years.
The working group brings together political scientists and national security experts versed in the geopolitics of AI alongside computer scientists with technical knowledge of the cutting edge of AI development.
The first project to receive joint funding through TPA and HAI, Human-Centered AI for Proliferation Monitoring, will see Mykel Kochenderfer of the School of Engineering’s Department of Aeronautics and Astronautics and Zegart, study how interpretable, human-centered AI systems can analyze large-scale, multimodal, open-source data to improve monitoring of nuclear proliferation. A multidisciplinary team will design AI-assisted workflows for two use cases: detecting illicit procurement for nuclear proliferation and identifying undeclared nuclear infrastructure around the world. The workflows draw on large language model–based AI agents and vision language models to integrate multilingual text, online multimedia, and satellite imagery.
For the second project, Public Opinion about AI and Geopolitics in the US and China, Michael Tomz of the School of Humanities and Sciences’ Department of Political Science will work alongside Diyi Yang of the School of Engineering’s Department of Computer Science. Their project addresses how AI is reshaping geopolitics in three ways: as something that confers military and economic advantage, as a weapon for information warfare, and as a source of new flashpoints such as Taiwan and its chip production. To assist, they will build CrossInterviewer, an AI-powered tool for adaptive interviews conducted in both Chinese and English, use it to survey US and Chinese citizens about AI and geopolitics, and analyze the results to inform academic and policy debates.
The third project, Political Agents: Chinese and US Foundation Models in Agent-Mediated Political Tasks, will see Jennifer Pan of the School of Humanities and Sciences’ Department of Communication and Sanmi Koyejo of the School of Engineering’s Department of Computer Science examine whether foundation-model provenance—meaning the training data, alignment choices, and regulatory regime behind a model—shapes what agent-mediated political systems produce.
Previous research shows that American and Chinese AI systems respond differently to sensitive topics. In practice, however, AI tools often work together in a series of steps, with each system using and building on the others’ responses. This project examines that process through the example of AI-assisted public comments on government regulations, where the technology is already being used and its outputs can be reviewed. The goal is to determine whether government agencies and political organizations should consider where an AI system comes from when deciding which tools to use.
Work on each project begins in August 2026 and runs until July 31, 2027.
For more information, please contact Jeffrey Marschner, assistant director of media and government relations, at jmarsch@stanford.edu or 202-760-3200.
