One warned that data centers will raise family electricity bills, the other pushed anti tariff framing built around Trump. Neither broke out, but both hit the same exposed seams in US AI policy.
OpenAI's June 10 Global Affairs threat report names two banned ChatGPT account clusters "likely originating from China" and frames them as covert attempts to manipulate US debate over AI and tech policy. The specific narratives those clusters pushed, on electricity prices and on tariffs, are not random. They track the two most contested fault lines in the current AI infrastructure fight: whether new data centers will raise costs for ordinary ratepayers, and how trade policy should treat the hardware that trains and runs frontier models.
OpenAI's threat report calls the first cluster "Data Center Bandwagon." According to the report, it generated social media comments and images arguing that AI data center buildouts are driving up electricity prices for average US families. The second cluster, "Tech and Tariffs," pushed an anti-tariff framing, with explicit instructions to feature Trump and exclude Xi, in OpenAI's telling. Both clusters were detected and banned before they had any visible traction.
That last point is the report's own caveat, and it deserves to lead. OpenAI says it found "no evidence of meaningful breakout" beyond the operations' own activity. Neither campaign reached a real audience, and OpenAI has not produced evidence that either narrative moved public opinion or reached a mainstream platform's recommendation system. The bans closed accounts, not influence operations that had already landed.
The more useful question is why these two narratives, and not others, were worth running through a generative model at all. Both target seams that already exist in legitimate US debate. Local opposition to data centers is a real, organized phenomenon: grid operators in Virginia, Texas, and the Midwest have spent the last year warning that AI demand will outpace available power, and at least a dozen state legislatures are weighing bills on siting, water use, and ratepayer protection. The "Data Center Bandwagon" content did not invent that concern. It attached itself to it.
The tariff story is the same shape in reverse. The Trump administration's tariff regime, including the export controls on advanced chips and the recent escalation around NVIDIA's H20, is a live policy fight with bipartisan friction. American AI labs, including OpenAI, have argued publicly that chip export controls risk ceding ground to Chinese competitors. A covert campaign pushing "stop the tariffs" framing to US consumers, with instructions to blame Trump and erase Xi, is an attempt to launder a contested industry preference into a populist consumer message. It is a recognizable influence pattern, even when the audience it reaches is zero.
The disclosing party here is also a stakeholder in both fights. OpenAI has signed its own long-term power purchase agreements, is building or planning data centers, and has lobbied on chip export rules. The threat report is a self-interested disclosure, even if the underlying finding is real. Readers should treat OpenAI's attribution and taxonomy as a company's claim about its own enforcement, not as a third-party forensic finding. OpenAI's "likely originating from China" language is the firm's attribution threshold, not an indictment.
The same caution applies to the choice to publish. Naming the campaigns gives OpenAI a public role in defining what counts as foreign manipulation of US AI policy debate, a debate the company is actively shaping through its own policy and product decisions. Threat reports from platforms are useful primary sources. They are also editorial acts.
What the report actually supports is narrower and more durable. Two clusters of accounts used ChatGPT to generate cheap content targeting the energy-cost and tariff stories. They were banned. They did not break out. The reported finding tells a reader where the exposed seams are in 2026 AI policy debate: anywhere a real local concern (electricity bills) or a real industry concern (chip access) can be packaged as a populist message. That pattern will outlast this report.