Pages

19 June 2025

New Report Shows How China Uses AI to Augment its Online Intelligence Operations


Artificial intelligence (AI) empowers China to conduct operations that blur the line between collecting sensitive information and influencing public opinion, and a June 10 report by OpenAI explains how.

Chinese intelligence operations are already leveraging online platforms to target government employees, military personnel, and academic researchers to recruit them as intelligence sources. OpenAI’s report shows how commercial AI tools can increase the speed, scale, and possibly the success of these operations.
Chinese Operation Blurs Line Between Influence and Intelligence

One operation captured in the OpenAI report specifically matches a known pattern of behavior from Beijing. As in many other Chinese campaigns, the operators pretended to be geopolitical risk consultants. They used OpenAI models to develop biographies for online personas, generate various types of content, analyze datasets, and translate correspondence.

While Chinese actors have used many of these tactics before, this particular operation is notable because those running it blended elements of intelligence gathering with foreign malign influence. The operators used OpenAI’s models to develop personas on X posing as journalists and geopolitical analysts who OpenAI assessed likely conducted a covert influence operation.

At the same time, the operators used OpenAI’s models to refine and translate unspecified correspondence intended for a U.S. senator about a government nominee. While OpenAI could not confirm whether the operators actually sent the correspondence, it demonstrates the intent to influence a member of Congress or at least gather intelligence from them.
Beijing Likely Uses Contractors to Conduct Espionage and Influence Operations

The prompts the operators used with OpenAI’s models suggest they were likely proxies or contractors rather than members of Beijing’s Ministry of State Security. OpenAI assessed that several prompts related to cyberattacks “suggest[ed] a low level of expertise.” These operators also used OpenAI’s models to create promotional materials aimed at the likely client for their services, the Chinese government.

No comments:

Post a Comment