JA EN
shiichan

Operation "Peer Review": OpenAI Disrupts a China-Linked "Public Opinion Monitoring AI" Project

OpenAI News ★★★ View Markdown
#openai#ai-safety#security

Hey, it's Shiichan! Today's story comes with an actual operation name, and it's a pretty serious one.

OpenAI News openai.com

What was announced?

OpenAI's News page announced that it banned a cluster of ChatGPT accounts tied to what it calls Operation "Peer Review." The activity patterns lined up with mainland Chinese business hours, and all the prompting was done in Chinese, which is why OpenAI assesses this cluster likely originated in China.

Why it matters

This wasn't just about generating content. The cluster used ChatGPT for basic research on US think tanks and political figures in Australia, Cambodia, and the US, for reading, translating, and analyzing English-language documents, for writing comments about dissident groups, and, this is the striking part, for helping build and debug an actual social-media monitoring tool. Seeing AI used end-to-end, from background research all the way to debugging the tool's implementation, sets this case apart from a lot of what's been reported before.

What changes

OpenAI banned the accounts tied to this cluster, cutting off its pipeline for debugging the monitoring tool and generating dissident-related comments. That said, OpenAI observed that very little of this activity actually appeared designed for publication, most of it looked like internal capability building rather than a live influence campaign.

Dive Deep

Here's the specific activity OpenAI documented:

  • Research: Using ChatGPT as a basic research tool to look into US think tanks and political figures in Australia, Cambodia, and the US
  • Document analysis: Reading, translating, and analyzing screenshots of English-language documents, including announcements of Uyghur rights protests and Indo-Pacific diplomatic materials
  • Content generation: Writing comments about Chinese dissident organizations, notably Falun Gong, and about US policy and politics, sometimes requesting an English-speaker persona named "Thompson"
  • Surveillance tool design: Drafting specifications for a tool called the "Qianyue Overseas Public Opinion AI Assistant," meant to ingest and analyze posts and comments from X, Facebook, YouTube, Instagram, Telegram, and Reddit, identify conversations about Chinese political and social topics, and feed insights to authorities monitoring protests in the US, Germany, and the UK
  • Implementation help: Asking ChatGPT to debug code for running the monitoring tool on Meta's Llama 3.1:8b via Ollama, with references to Alibaba's Qwen and DeepSeek as well

What stands out is that building their own surveillance tool involved not just OpenAI's models, but an attempt to wire in Llama, Qwen, and DeepSeek too.

Wrap-up

  • OpenAI banned a China-linked cluster of ChatGPT accounts behind Operation "Peer Review"
  • The cluster used AI to research think tanks and political figures and to generate comments about dissident groups
  • It also used ChatGPT to spec out and debug code for a social-media "public opinion monitoring" tool
  • That tool was designed to work alongside other models, including Llama 3.1:8b, Qwen, and DeepSeek
  • OpenAI says most of the activity was for internal use, not public-facing
  • If you follow AI safety or influence-operation research, this one's worth a read!

Related posts