Chinese law enforcement used ChatGPT and Chinese open-weight models in a large-scale influence operation targeting dissidents and foreign political figures, including the Japanese Prime Minister, as revealed by OpenAI in their latest threat report.
Chinese law enforcement used ChatGPT and Chinese open-weight models in a large-scale influence operation targeting dissidents and foreign political figures, including the Japanese Prime Minister, as revealed by OpenAI in their latest threat report.
Scale: The activity involved hundreds of operators and thousands of fake accounts. Targets included Japan’s PM Sanae Takaichi, as well as prominent dissidents and human rights group Safeguard Defenders. Some dissident accounts appear to have been intimidated by harassment, but the campaign’s overall impact was relatively small.
Methods: ChatGPT, which initially refused assistance with operational planning, was primarily used to refine internal reports and plans and improve messaging. The operation also seemingly relied on locally deployed open-weight models, including DeepSeek-R1 and Qwen2.5, for social media monitoring, translation, content generation, and workflow automation.
Attribution: The operation is tied to Chinese law enforcement and aligned with Spamouflage, a long-running Chinese IO targeting global audiences that has been previously linked to attempts to use ChatGPT and Gemini.
