OpenAI has banned a cluster of ChatGPT accounts, believed to originate from China, that were using its models to aid in surveillance planning. Dubbed "Peer Review," the operation involved analyzing documents, generating sales pitches for social media monitoring tools, debugging code, and researching political figures.

The network's activity patterns, including prompts in Chinese and operating during Chinese business hours, indicated a connection to mainland China. The operators used ChatGPT as a research tool, analyzed English-language documents potentially related to Uyghur protests and Indo-Pacific diplomacy, and generated comments on Chinese dissident groups and U.S. policies.

A significant focus was on developing and promoting an AI-powered social media listening tool called the "Qianyue Overseas Public Opinion AI Assistant." Descriptions suggest this tool was intended to monitor platforms like X and Facebook, identify conversations on Chinese political topics, and report insights to Chinese authorities. The operators also used ChatGPT to proofread claims of sending insights to Chinese embassies and intelligence agents. Code debugging for this tool, referencing models like Meta's llama3.1:8b, was also observed.

While much of the activity focused on capability development and internal use, OpenAI notes that the full impact requires further investigation by stakeholders. The operation also involved debugging code for malware analysis and generating phishing emails, though the latter was not confirmed to be AI-powered.