Microsoft on Thursday accused Chinese operatives of using AI to sow discord among US voters.
Budrul Chukrut/SOPA Images/LightRocket via Getty Images
Microsoft warned that China has been using AI-generated images to sow discord among US voters.
Real people are reposting these divisive images, not realizing they’re fake, Microsoft added.
The Chinese Embassy in Washington, D.C., hit back, speaking of “malicious speculation” against China.
China has since March been using generative AI to create viral content that divides US voters on social media — and is getting better at it, Microsoft analysts say.
“We have observed China-affiliated actors leveraging AI-generated visual media in a broad campaign that largely focuses on politically divisive topics, such as gun violence, and denigrating US political figures and symbols,” wrote Clint Watts, general manager of the Microsoft Threat Analysis Center.
Analysts from the center released a report on Thursday highlighting digital threats from China and North Korea, and said Chinese operatives are influencing their targets to “a greater extent than previously observed.”
Microsoft warned of social media accounts affiliated with the Chinese Communist Party, which would post AI-generated images like a Black Lives Matter poster showing bullet holes and a Black man surrendering.
Hours after the Black Lives Matter poster was uploaded, an account impersonating a conservative US voter reposted the image, complaining about racial discrimination, according to Microsoft.
Another cited example was an AI-generated poster of the Statue of Liberty holding an assault rifle, along with the caption “The Goddess of Violence.” Microsoft analysts said it was posted by an account suspected of being run as a Chinese influence operation.
“This relatively high-quality visual content has already drawn higher levels of engagement from authentic social media users,” the analysts warned.
They added that the new images are more eye-catching, and real US voters are reposting them even though they contain “common indicators of AI-generation,” like more than five fingers on a hand.
It’s now also harder to tell that these accounts aren’t genuine, Microsoft’s analysts said.
“Unlike earlier IO campaigns from CCP-affiliated actors that used easy-to-spot computer-generated handles, display names, and profile pictures, these more sophisticated accounts are operated by real people who employ fictitious or stolen identities to conceal the accounts’ affiliation with the CCP,” they wrote.
But there are tell-tale signs, according to Microsoft. Many of these operatives would start posting in Mandarin, then switch to another language, the threat center said.
The accounts would make a post, then comment, like, and share posts from other similar accounts, creating a pattern where they would boost each others’ content, Microsoft said.
“We can expect China to continue to hone this technology over time, though it remains to be seen how and when it will deploy it at scale,” Watts wrote.
When asked to respond to Microsoft’s report, a spokesperson for the Chinese Embassy in Washington, D.C., Liu Pengyu, told Insider in a statement: “In recent years, some western media and think tanks have accused China of using artificial intelligence to create fake social media accounts to interfere the US’ politics. Such remarks are full of prejudice and malicious speculation against China, which China firmly opposes.”
In July, cybersecurity research firm Mandiant said that it found Chinese-affiliated operatives had paid people in Washington, D.C., to protest a US government ban on goods produced in the Xinjiang region. Videos of their protests were then “amplified” by social media accounts used by the operatives, Mandiant added.
On August 29, Meta also warned of a Chinese influence campaign spreading false information, involving more than 7,700 Facebook accounts, 950 Facebook pages, 15 Facebook groups, and 15 Instagram accounts. The social media company said it removed all of them.