OpenAI banned PRC-linked accounts building a 'High-Risk Uyghur' surveillance model

OpenAI has banned a cluster of accounts linked to individuals apparently working for Chinese government entities, revealing a pattern of using ChatGPT not for grand-scale AI deployment, but for the bureaucratic planning of surveillance and targeted profiling. The activity, detailed in a threat intelligence report, is a stark reminder that authoritarian abuse of AI often starts with the paperwork. The banned users weren’t using the model to directly spy on people. Instead, they were asking it to help design the tools for the job, like drafting a project proposal for a social media ‘probe’ that could allegedly scan Western platforms for dissident content, and write promotional materials for a system pitched to Chinese security forces. One particularly chilling case involved a user seeking help to write a proposal for a ‘High-Risk Uyghur-Related Inflow Warning Model’ that would analyze transport bookings against police records to track the movements of a vaguely defined category of people. This wasn’t a sophisticated, institutional adoption of AI; it was desk-level work, consistent with individual officials trying to accelerate their own workflows. The firm also disrupted accounts conducting targeted research on government critics, including attempts to identify the funding sources for an anti-government X account and the organizers of a petition in Mongolia. In these cases, the model only surfaced publicly available information. The findings support OpenAI’s broader argument for ‘democratic AI’ governance, where preventing authoritarian regimes from using AI to consolidate power is a core design principle, not an afterthought.