Fake Accounts Using GAN Faces Deployed by Propaganda Campaign on Social Platforms

A pro-China propaganda campaign dubbed 'Spamouflage Dragon' by research company Graphika operated from June to early August 2020, posting English-language videos critical of the Trump administration on Facebook, Twitter, and YouTube. The campaign used AI-generated images, likely created by generative adversarial networks (GANs), to create fake profile photos for social media accounts. Graphika identified these fake images by spotting commonalities such as blurred backgrounds and aligned eyeball positions. The AI-generated photos were used to create fake followers on Twitter and YouTube to amplify the propaganda content. The videos themselves were poorly made with language errors and awkward automated voice-overs, including mispronunciations like 'us' instead of 'US' and grammatical errors such as 'Public blamed Trump sinaction.' The campaign targeted Trump's TikTok ban and COVID-19 response but failed to receive engagement from real users. Social media companies eventually removed the videos and affiliated accounts. This marked the first time Spamouflage Dragon, which had previously focused on Chinese-language content supporting the CCP, expanded to substantial English-language content targeting US audiences and used AI-generated profile pictures.