Blab AI
Stanford HAI and Hoover Institution Award $300,000 in Grants to Study AIs Geopolitical Impact
In April 2026, Stanford’s Institute for Human‑Centered Artificial Intelligence (HAI) and the Hoover Institution’s Technology Policy Accelerator announced a new seed‑grant program that will fund three multidisciplinary research teams. Each team will receive $100,000 to investigate how foundation models, agents, and other AI systems affect global security, public opinion, and political influence. The initiative reflects a growing consensus that artificial intelligence is reshaping geopolitics and that policymakers need evidence‑based guidance.
The grant call required teams to combine a technical lead with a social‑science or humanities scholar. According to the program’s announcement, the goal is to “catalyze original research that helps policymakers understand where the greatest risks and opportunities lie.” Amy Zegart, a senior fellow at the Hoover Institution, said the program would bring together experts across campus to examine nuclear security, public opinion, and AI’s role in political decision‑making. Chris Manning, a Stanford professor of linguistics and computer science, added that the projects would help leaders anticipate and respond to AI’s geopolitical impact.
The first selected project tackles nuclear proliferation. Associate Professor Mykel Kochenderfer of Aeronautics and Astronautics and HAI Senior Fellow, together with Amy Zegart, will develop AI agents that act as “digital detectives.” The agents will process multilingual documents, still images, video, and satellite photos to detect changes that could signal illicit nuclear development. The team will also create a standardized benchmark to evaluate the reliability of the system. The research builds on prior work that showed AI‑assisted workflows can identify nuclear proliferation signals in large satellite imagery datasets.
The second project addresses public attitudes toward AI in the United States and China. Political‑science professor Michael Tomz and computer‑science assistant professor Diyi Yang will build CrossInterviewer, a bilingual AI‑powered interviewing tool. The system will conduct interviews in Chinese and English, adapting survey questions based on each respondent’s answers. The researchers plan to examine three themes: general attitudes toward AI, willingness to adopt AI at work, and preferences for open versus closed foundation models. By capturing nuanced, in‑depth responses at scale, the team aims to provide a firmer basis for government policy, corporate investment, and forecasts about AI’s future.
The third project explores how the provenance of foundation models influences political outcomes. Professor Jennifer Pan of communication and associate professor Sanmi Koyejo of computer science will analyze closed U.S.–China trade dockets as a test case. They will examine how agent teams built on models from different countries produce summaries and how those summaries affect policy decisions. The researchers will release a public dataset of political‑task agents with their underlying models identified, offering the first look at which foundation models are actually being deployed in political applications.
Collectively, the projects aim to fill critical knowledge gaps. The nuclear‑proliferation study could provide analysts with new tools to monitor illicit activity in real time. The public‑opinion research will clarify why enthusiasm for AI differs between the U.S. and China and how workers in both countries view workplace automation. The provenance study will shed light on whether the origin of a model can bias political decision‑making, a question that has become urgent as governments deploy agents built on open‑weight Chinese models.
The grants are scheduled to fund research over the next year. While the projects are still in early stages, the work will likely produce policy briefs, datasets, and prototype tools that could inform regulators, defense analysts, and corporate leaders. Unresolved questions remain about how to standardize benchmarks for AI‑driven proliferation detection, how to ensure CrossInterviewer’s linguistic accuracy across diverse dialects, and how to enforce provenance disclosure in government and political applications. Nonetheless, the initiative represents a concrete step toward translating technical AI advances into actionable geopolitical insights.
The grant call required teams to combine a technical lead with a social‑science or humanities scholar. According to the program’s announcement, the goal is to “catalyze original research that helps policymakers understand where the greatest risks and opportunities lie.” Amy Zegart, a senior fellow at the Hoover Institution, said the program would bring together experts across campus to examine nuclear security, public opinion, and AI’s role in political decision‑making. Chris Manning, a Stanford professor of linguistics and computer science, added that the projects would help leaders anticipate and respond to AI’s geopolitical impact.
The first selected project tackles nuclear proliferation. Associate Professor Mykel Kochenderfer of Aeronautics and Astronautics and HAI Senior Fellow, together with Amy Zegart, will develop AI agents that act as “digital detectives.” The agents will process multilingual documents, still images, video, and satellite photos to detect changes that could signal illicit nuclear development. The team will also create a standardized benchmark to evaluate the reliability of the system. The research builds on prior work that showed AI‑assisted workflows can identify nuclear proliferation signals in large satellite imagery datasets.
The second project addresses public attitudes toward AI in the United States and China. Political‑science professor Michael Tomz and computer‑science assistant professor Diyi Yang will build CrossInterviewer, a bilingual AI‑powered interviewing tool. The system will conduct interviews in Chinese and English, adapting survey questions based on each respondent’s answers. The researchers plan to examine three themes: general attitudes toward AI, willingness to adopt AI at work, and preferences for open versus closed foundation models. By capturing nuanced, in‑depth responses at scale, the team aims to provide a firmer basis for government policy, corporate investment, and forecasts about AI’s future.
The third project explores how the provenance of foundation models influences political outcomes. Professor Jennifer Pan of communication and associate professor Sanmi Koyejo of computer science will analyze closed U.S.–China trade dockets as a test case. They will examine how agent teams built on models from different countries produce summaries and how those summaries affect policy decisions. The researchers will release a public dataset of political‑task agents with their underlying models identified, offering the first look at which foundation models are actually being deployed in political applications.
Collectively, the projects aim to fill critical knowledge gaps. The nuclear‑proliferation study could provide analysts with new tools to monitor illicit activity in real time. The public‑opinion research will clarify why enthusiasm for AI differs between the U.S. and China and how workers in both countries view workplace automation. The provenance study will shed light on whether the origin of a model can bias political decision‑making, a question that has become urgent as governments deploy agents built on open‑weight Chinese models.
The grants are scheduled to fund research over the next year. While the projects are still in early stages, the work will likely produce policy briefs, datasets, and prototype tools that could inform regulators, defense analysts, and corporate leaders. Unresolved questions remain about how to standardize benchmarks for AI‑driven proliferation detection, how to ensure CrossInterviewer’s linguistic accuracy across diverse dialects, and how to enforce provenance disclosure in government and political applications. Nonetheless, the initiative represents a concrete step toward translating technical AI advances into actionable geopolitical insights.
