Operation “Peer Review”: AI-assisted surveillance planning

This case study was originally published in OpenAI’s February 2025⁠(opens in a new window) report.

Summary

Actor

We banned a cluster of ChatGPT accounts that, based on behavioral patterns and other findings, likely originated in China. They were using our models to assist with analyzing documents, generating sales pitches and descriptions of tools for monitoring social media activity powered by non-OpenAI models, editing and debugging code for those tools, and researching political actors and topics. Based on this network’s behavior in promoting and reviewing surveillance tooling, we have dubbed it “Peer Review.”

Behavior

This network consisted of ChatGPT accounts that operated in a time pattern consistent with mainland Chinese business hours, prompted our models in Chinese, and used our tools with a volume and variety consistent with manual prompting, rather than automation. In one instance, we believe the same account may have been used by multiple operators.

Within the cluster, the operators performed a number of primary tasks. One was to use the model as a basic research tool, in a way similar to which earlier generations would have used search engines. This included, for example, searching for publicly available information about think tanks in the United States, and politicians and government officials in countries including Australia, Cambodia, and the United States.

Another workstream consisted of using our models to read, translate, and analyze screenshots of English-language documents. Some of these images were announcements of Uyghur rights protests in a range of Western cities, and were potentially copied from social media. Others appeared to concern diplomatic and government topics in the Indo-Pacific region. There is insufficient evidence to determine whether these documents were authentic, or to show how the operators obtained them.