Natalie Winters@nataliegwinters · Aug 24, 2026

What would China need to influence an American election? • Find voters who can be moved • Test which messages move…

9 tweets1 min read♥ 3Koriginal

Summary

Chinese state universities have developed AI capabilities for analyzing and influencing American elections, including systems to generate political replies, identify swing voters, simulate audience reactions to political content, and predict individual user behavior. These projects were funded by China's national science programs and conducted at universities formally connected to the Chinese government.

Summarized by ThreadOut AI from the full thread. May miss nuance — read the thread below.

#1
What would China need to influence an American election? • Find voters who can be moved • Test which messages move them • Predict how people will react • Mass-produce political replies Chinese state universities have now built each capability. The documents: 🧵
#2
During the 2024 race, researchers at Beijing University of Posts and Telecommunications collected 1,039 Trump/Harris posts and 25,025 replies. They trained AI to automatically write either a supportive or hostile reply to a candidate’s post. arxiv.org/pdf/2504.03295
#3
The AI followed the requested political stance 92.57% of the time. Its authors said it could help “shape public opinion.” Translation: automated pro- or anti-candidate replies. BUPT is a state university co-built with China’s technology ministry. zsc.bupt.edu.cn/info/1016/1230…
#4
At Beihang University, researchers mapped 323,105 accounts in a U.S.-election network and inferred their political leanings. Then they modeled how targeted content could change swing-voter behavior—and potentially the election result. journals.aps.org/prresearch/pdf…
#5
This wasn’t just an effort to count Republicans and Democrats. The project modeled which politically movable users could be pushed with targeted content. Its public code repository is literally titled “Manipulation of Swing Voters.” github.com/ziqianliu2001-…
#6
Fudan University built what amounts to an AI political focus group using nearly 10,000 real X users. It profiled age, gender, income, education, employment, ideology, party support and “election tendencies.” aclanthology.org/2025.coling-de…
#7
A user can choose the political makeup of an audience, feed it a news story, and watch AI simulate how that audience will react—before publishing it. Researchers tested the system on coverage questioning Trump’s age and fitness for office. aclanthology.org/2025.coling-de…
#8
Tsinghua researchers went further: they built AI “role-playing agents” from 1.7 million tweets surrounding the 2020 election. The system studies someone’s prior posts, tone, attitudes and followed accounts—then predicts that user’s next political reaction. aclanthology.org/2025.acl-long.…
#9
These were not independent hobby projects. China’s national science programs funded the Beihang, Fudan and Tsinghua work. BUPT and Beihang are formally tied to China’s Ministry of Industry and Information Technology.