OpenAI: Chinese agent used ChatGPT for smear ops

"Despite the name, these are basically operations designed to harass and silence domestic and foreign critics, closer to what we'd think of as covert influence operations and transnational repression. The malicious activity OpenAI spotted included generating status reports on operations targeting Chinese dissidents and CCP critics, along with the specific covert op against Takaichi."

"The user, since banned by the AI giant, tried to convince the model to help them plan a smear campaign against Sanae Takaichi, Japan's first female prime minister, after she criticized the Chinese Communist Party for trampling human rights in Inner Mongolia. Some of the prompts included asking the model to post and amplify negative comments about her on social media."

"The latter seemed to follow the structure of the original draft plan to discredit the Japanese politician, focusing on the same five areas - negative comments, immigration, living conditions, far-right links, and tariffs - and even provided the threat intel team with several operational details about the smear campaign."

OpenAI discovered a user with connections to Chinese law enforcement attempting to use ChatGPT for coordinated smear campaigns targeting Japanese Prime Minister Sanae Takaichi and critics of the Chinese Communist Party. The user requested assistance in planning negative social media campaigns, creating fake email accounts, and amplifying criticism against Takaichi following her human rights criticisms of the CCP in Inner Mongolia. When ChatGPT refused, the user reportedly employed other AI models to execute the campaign. Subsequently, the same user asked ChatGPT to edit status reports on cyber operations targeting Chinese dissidents and CCP critics. OpenAI's investigation classified these activities as covert influence operations and transnational repression designed to harass and silence both domestic and foreign critics.

Read at Theregister