OpenAI Uncovers Chinese Influence Operations Using ChatGPT to Manipulate Data Center Debate

OpenAI's threat intelligence team has identified two distinct online activity clusters with suspected ties to China, which utilized ChatGPT to generate content designed to inflame existing domestic debates. The first operation, dubbed "Data Center Bandwagon," focused on creating social media posts and imagery alleging that the buildout of data centers was increasing electricity prices for Americans. This campaign sought to tap into pre-existing concerns about energy costs and infrastructure development.

The second operation employed ChatGPT to produce content criticizing tariffs, specifically directing the AI to focus on U.S. President Donald Trump while omitting Chinese President Xi Jinping. This suggests an attempt to frame tariff discussions in a way that aligned with specific geopolitical narratives. Both operations posed as ordinary Americans on social media platforms like X and YouTube, using simplified Chinese prompts and requesting both English and Chinese-language outputs, according to OpenAI.

Researchers noted that these campaigns appear to be a classic example of foreign influence operations attempting to manipulate genuine domestic debates. Ben Nimmo, a principal investigator at OpenAI, stated that the operators "likely originated" in China. The anti-data center content was traced to an unnamed Chinese technology company with contracts from regional Chinese governments. Both clusters utilized VPNs to evade detection and employed tactics to maintain persistent and credible online accounts, produce visually appealing content, and anticipate platform enforcement.

Despite the sophisticated use of AI tools, the report indicates that these influence campaigns had limited success in gaining traction beyond their own amplification networks. OpenAI rated the campaigns low on their "Bookings breakout scale," signifying activity on platforms but no meaningful engagement from targeted audiences. This suggests that while AI can scale content creation, it does not guarantee influence or widespread adoption of the messaging.

Experts in state-sponsored influence campaigns observe that such groups often latch onto and amplify existing domestic movements or messaging to serve their broader destabilization goals. By piggybacking on established narratives, like public anger over AI and data centers, these operations can create the illusion of greater effectiveness. However, the content generated sometimes appeared clunky or overly direct, betraying a lack of familiarity with English language nuances and internet virality.

Nimmo emphasized that these operations did not create the debates but rather attempted to interfere in existing ones. "We didn’t see any signs that it succeeded," he stated, acknowledging that while the views expressed were reasonable and sincerely held by many, the covert foreign manipulation was a concern. The actors even used ChatGPT to refine operational security details for their social media campaigns and to strategize on leveraging platform features like Facebook's content ecosystem and advertising tools.

Furthermore, the campaign targeting tariffs also used ChatGPT to generate comments and comics in multiple languages, accusing the U.S. of prioritizing profits over alliances. OpenAI also reported being targeted by the same network with a false influence campaign alleging a widespread user data breach. While OpenAI stopped short of directly attributing the operations to the Chinese government, the tactics and campaign elements strongly overlap with established Chinese government propaganda efforts online.

This incident highlights the evolving tactics of influence operations, which are increasingly leveraging advanced AI tools like ChatGPT to generate and disseminate propaganda at scale. The campaigns underscore the challenges in distinguishing genuine domestic discourse from sophisticated foreign manipulation, especially when AI is used to mimic human behavior and exploit existing societal divisions.