This week, Natalie Winters published what might be the most important AI-geopolitics investigation of the year. The headline sounds like a Black Mirror pitch — "China Is Building AI Models of American Voters" — but the sourcing is concrete, and the implications are chilling.
Fudan University built ElectionSim: a system that scraped 171 million X posts, classified users by age, gender, race, ideology, and party affiliation, then simulated presidential elections across all 50 states. Their model reproduced the winner in 47 of 51 state-level contests. But it doesn't stop at prediction — the system lets operators select voters by demographic profile and conduct multi-round conversations with synthetic constituents. Synthetic focus groups derived from real social-media behavior.
The Shanghai Academy of Social Sciences went further. They ran a controlled political-message experiment on 300 synthetic Pennsylvania voters — introducing a candidate, measuring shifts in stated concerns, and using treatment/control conditions to extract usable political intelligence.
Shanghai Jiao Tong University published studies using ANES and Chicago Council survey data to model "America First" attitudes toward tariffs, alliances, and international organizations — explicitly framed as a tool for shaping Chinese foreign policy toward the U.S.
The pattern is clear. This isn't academic curiosity. These are state-funded, policy-directed projects building digital twins of the American electorate — not to study, but to simulate, test, and exploit.
Independent verification is limited because the papers are in Chinese. But the abstracts, author affiliations (state think tanks), and explicit policy language are documented. If even half of this is operationally real, someone in the U.S. intelligence community should be very awake right now.