Overview
- OpenAI’s Feb. 25 threat report details a banned ChatGPT account whose activity outlined a large, sustained program to suppress critics at home and abroad.
- The user tried to enlist ChatGPT to plan an operation against incoming Japanese prime minister Sanae Takaichi; the model refused, and later uploads indicated the campaign proceeded using other tools.
- Investigators matched the uploads to real‑world activity, including a 2023 false obituary, forged documents, impersonation of U.S. officials, and mass reporting that contributed to dissident account restrictions.
- The operation described relied on hundreds of personnel, thousands of fake accounts across hundreds of platforms, and locally deployed AI models such as DeepSeek and Qwen, with ChatGPT used mainly for planning and editing rather than hacking.
- OpenAI also reported separate abuses in the same cycle, including a Cambodia‑linked romance scam likely defrauding hundreds monthly, clusters posing as law firms or law enforcement, and a China‑linked cluster seeking information on U.S. persons and federal building locations.