Chinese Influence Network Used A.I. to Impersonate U.S. Voters

Microsoft on Thursday said it has detected a Chinese-controlled network of social media accounts that uses A.I. technology to impersonate American voters and spread propaganda to influence U.S. politics.

Microsoft’s report said the suspected Chinese influence network was similar to past operations that have been linked to the Chinese Ministry of Public Security. 

The difference with the new network is that it began using generative A.I. in March 2023 to “mimic U.S. voters” and produce content that was “more eye-catching than the awkward visuals used in previous campaigns by Chinese nation-state actors, which relied on digital drawings, stock photo collages, and other manual graphic designs.”

“These images are most likely created by something called diffusion-powered image generators that use A.I. to not only create compelling images but also learn to improve them over time,” Microsoft said.

The influence network engaged in “a broad campaign that largely focuses on politically divisive topics, such as gun violence, and denigrating U.S. political figures and symbols.”

An example presented by Microsoft researchers was an A.I.-generated image of the Statue of Liberty brandishing an assault rifle over a caption that said, “Everything is being thrown away. THE GODDESS OF VIOLENCE.” 

Another example was a Black Lives Matter poster that was evidently created by a “Chinese Communist Party-affiliated automated account,” and then “uploaded by an account impersonating a U.S. conservative voter seven hours later.”

These social media post do not appear terribly persuasive, nor does their origin as Chinese propaganda seem very well-disguised, but Microsoft’s point was that artificial intelligence generated the post with minimal human intervention and is learning from its mistakes, building a more formidable propaganda operation as it moves forward. Researchers noted the Chinese influence networks are getting better at responding to human interaction.

A Microsoft spokesperson told Reuters on Thursday that the China-linked accounts “attempted to appear American by listing their public location as within the United States, posting American political slogans, and sharing hashtags relating to domestic political issues.”

Microsoft’s threat report made some other general observations about Chinese cyber activity, including “the South China Sea, the U.S. defense industrial base, and U.S. critical infrastructure” as the top three subjects of Chinese hacking activity in 2023.