OpenAI Reports China Employing AI-Driven Surveillance Technologies

OpenAI’s latest threat report reveals that Chinese influence operations are increasingly leveraging artificial intelligence to conduct surveillance and disinformation campaigns. These tactics highlight a growing concern for businesses regarding the use of advanced technologies in cyber operations.

The report identifies two notable campaigns that demonstrate the misuse of AI tools, including OpenAI’s models, to further state-sponsored objectives. One of these, referred to as “Peer Review,” involved the development of an AI-driven social media monitoring tool aimed at tracking anti-China sentiments in Western nations.

Researchers from OpenAI discovered this operation when they observed an individual utilizing the company’s AI technology to debug code for the surveillance tool. This case marks a significant moment in cybersecurity history, noted Ben Nimmo, a principal investigator at OpenAI, as it represents one of the first instances of an AI-enhanced surveillance mechanism being publicly identified. The tool is believed to be built upon Meta’s open-source AI model, Llama.

The functionality of the surveillance tool reportedly includes real-time reporting on protests and dissident activities, with intelligence being relayed to Chinese security agencies. In response, OpenAI has banned accounts affiliated with this project, asserting that such applications violate its policies against AI-enabled unauthorized surveillance.

Further, OpenAI outlined another campaign dubbed “Sponsored Discontent,” which utilized AI to propagate anti-U.S. narratives within Spanish-language media. This initiative produced and translated articles critical of American society and politics, distributing them across Latin American media outlets, often under the guise of sponsored content. This operation also incorporated automated English-language social media comments aimed at discrediting Chinese dissident Cai Xia.

Nimmo pointed out that this instance represents the first known effort by Chinese influence operatives to systematically produce and distribute long-form articles in Spanish for a Latin American audience. Without OpenAI’s analytical visibility into its models, linking the social media activities to broader media efforts would have been exceedingly challenging.

The report also draws attention to other AI-related cyber threats, including scams and influence operations attributed to North Korea and Iran, as well as election interference in Ghana. These findings suggest that the evolution of open-source AI models—which can now be deployed locally—will complicate detection and response to such nefarious uses.

In the “Peer Review” case, references to popular AI tools like ChatGPT, DeepSeek, and Meta’s Llama 3.1 indicate that operators may be experimenting with various models to obscure their initiatives. This emphasizes the importance of vigilance and preparedness in the cybersecurity landscape, as businesses must remain alert to the potential misuse of advanced technologies that can compromise their security posture.