OpenAI published a case study, first released in its June 2026 threat report, on a banned cluster of ChatGPT accounts that likely originated in China. The accounts prompted in Simplified Chinese over VPNs and generated short English comments and cartoons criticizing US tech policy and tariffs, which likely inauthentic X accounts then posted. The operators also generated content in Italian, Japanese, and Traditional Chinese.
The case shows a model provider attributing an influence operation by joining its own prompt records with behavioral signals observed on another platform. The prompts supplied content and intent, including requests for bulk comments with short character limits, while reposts and shared quote-tweets on X tied the content accounts to the separate data-compromise cluster. OpenAI also reports that when the operators asked for a concept for an online public-opinion surveillance system, the model gave general data storage advice but no collection methods.
The timing matters because OpenAI itself was a target, and the campaign followed a CCP plan that elevated AI as a strategic priority. For trust and safety teams, cross-platform account behavior carried more attribution weight here than any stylistic trace of generated text.
The report does not establish audience impact, since OpenAI rated the activity Category One on the Breakout Scale with little observable engagement. OpenAI also says it could not establish the operators' precise institutional affiliation. Further evidence would include X or independent researchers confirming the network links, or any sign of breakout beyond one platform.