OpenAI Says China-Linked Operatives Used ChatGPT to Stoke US Tariff and Data Center Debates

OpenAI says it caught and shut down a set of accounts linked to Chinese influence operatives who used ChatGPT to generate content aimed at US debates over tariffs and the rapid construction of AI data centers, according to a report surfaced by Axios. The activity targeted two of the most economically loaded fault lines in current US policy, where trade costs and infrastructure spending intersect with national security.

The pattern matters because it shows how the tools being sold as productivity engines are also being repurposed as content factories for foreign influence work. OpenAI has built a recurring threat-intelligence practice around exactly this, publishing periodic reports on how state-linked actors attempt to use its models. The company's usage policies prohibit covert influence operations, and its safety and threat disclosures have become a regular channel for documenting takedowns. The latest disclosure puts tariffs and data center expansion in the crosshairs, two subjects where US public opinion carries direct commercial weight.

What the operatives were after

Tariffs and data centers are not random targets. Both touch the cost structure of the American technology sector and the broader question of how much the US should spend, and where, to keep an edge in artificial intelligence. Influence content that nudges public sentiment on tariff policy can affect the price of imported hardware, the components that flow into servers, and the calculus companies use when deciding to build domestically or abroad. Content that frames AI data centers as environmentally reckless or fiscally wasteful can slow permitting, raise local opposition, and complicate the financing of projects that already run into the billions.

The scale of the data center question is the part that gives this campaign its commercial logic. US hyperscalers including Microsoft, Amazon, Google, and Meta have collectively guided toward capital expenditure well above $300 billion for the current year, with the majority earmarked for AI compute and the facilities to house it. Each large training cluster represents a multibillion-dollar bet, and each one depends on local approvals, grid capacity, and a political environment willing to tolerate the power draw. An influence operation that shifts even a fraction of public opinion in key jurisdictions operates on a favorable cost-to-impact ratio, since the content is cheap to produce and the projects it can delay are enormously expensive.

Why model providers keep finding this

OpenAI is not unique in reporting state-linked misuse. The detection itself has become a competitive and reputational asset. By publishing takedowns, model providers signal to regulators and enterprise customers that they monitor abuse, which matters as governments weigh how tightly to oversee frontier AI. Anthropic, Google, and Meta have all released similar threat findings tied to their own systems. The reporting cadence has effectively turned the major labs into an informal intelligence apparatus, mapping how influence actors adopt generative tools.

The business read here is that content generation costs have collapsed. Producing persuasive, fluent text at volume once required staff and time. A language model compresses that into near-zero marginal cost, which lowers the barrier to running a sustained campaign and raises the volume of material defenders have to sift through. The operatives in this case appear to have used ChatGPT as a drafting engine rather than a distribution network, generating posts and arguments that were then pushed out through other channels.

What changes from here

The immediate consequence is more pressure on platforms and providers to share signals. Detecting that text was machine-generated is hard once it leaves the model and lands on a social network, so the most reliable point of interdiction is the account level inside the AI provider itself, exactly where OpenAI acted. That argues for tighter coordination between the labs and the platforms where influence content ultimately surfaces.

For companies building the data centers, the takeaway is that local opposition can no longer be assumed to be purely organic. Project developers planning large compute campuses will increasingly need to account for synthetic amplification when they gauge community sentiment, and policymakers weighing tariff changes face a discourse that can be seeded at scale by foreign actors. The economics that make AI compute valuable, low marginal cost and high leverage, are the same economics that make it attractive to the people trying to shape the politics around it. OpenAI's disclosure is a reminder that the infrastructure debate and the information war over it are now running on the same rails.