OpenAI bans China-linked accounts running influence ops through ChatGPT

OpenAI disclosed Wednesday that it has shut down China-linked accounts using ChatGPT to generate social media content aimed at shaping American public opinion on tariffs and AI data center policy. The company identified two separate operations, including one labeled "Data Center Bandwagon" that produced comments and political cartoons designed to amplify divisions in U.S. tech policy debates.

The campaigns appear to have fallen flat. OpenAI noted they showed no meaningful engagement or audience reach. But the disclosure matters less for what the operations achieved than for what they reveal about the experimental phase now underway. State-aligned actors are stress-testing generative AI as a content mill for influence work, probing where automation can replace or scale human effort in narrative campaigns.

The timing is notable. Tariff policy and data center infrastructure sit at the center of U.S.-China economic competition, and both debates carry direct implications for capital allocation, supply chain posture, and regulatory risk. An influence operation that moves sentiment even incrementally could shift legislative appetite or public tolerance for restriction. So far, these efforts have not demonstrated that capability, but the low cost of production means iteration is cheap.

The FBI separately announced it had seized 13 websites allegedly used by China to recruit U.S. workers, according to the Associated Press, underscoring a broader pattern of digital infrastructure being deployed for intelligence and influence purposes. OpenAI's disclosure fits within that wider operational landscape, where the boundary between recruitment, espionage, and narrative shaping increasingly blurs.

For stewards watching platform governance and geopolitical risk, the lesson is not about this campaign's impact but about the rehearsal. Influence operations are moving into the same generative stack that companies are adopting for customer service, marketing, and content production. The tools are neutral, the economics are favorable, and the experimentation is live.