OpenAI says China-linked bots used ChatGPT to attack US data centers

OpenAI says it found a suspected China-linked influence operation that tried to sway the US data center debate by using ChatGPT. OpenAI CEO Sam Altman has not been cited here directly, but the company’s Intelligence and Investigations team did: Ben Nimmo, principal investigator on OpenAI's Intelligence and Investigations team, told reporters on Wednesday ahead of the report’s release that the tactic looked like a foreign influence operation “jumping onto the bandwagon of a genuine pre-existing domestic debate.” The target was the hot-button reality of data centers, which OpenAI and its competitors describe as central to cementing the United States’ lead in the global AI race and meeting current and expected future demand.

According to OpenAI, the campaign was identified because the users relied in part on ChatGPT, and it did not gain meaningful traction online. OpenAI described the operation, which it dubbed the “Data Center Bandwagon” Campaign, as small scale and short-lived. It said most posts gained little to no authentic engagement on X and also Facebook, and it added that it found no evidence of meaningful breakout beyond the operation’s own activity. Nimmo said there were “no signs that they succeeded.” In plain English: the play may have been sophisticated enough to show up, but not effective enough to substantially change the public conversation.

So why does this matter beyond the obvious “foreign influence is bad” headline? Because OpenAI says the campaign used the exact kind of tools that have become infrastructure for modern discourse. The report says the users behind the data center effort “were likely part of a social media operations team at a private Chinese technology company conducting work for Chinese provincial-level government clients.” OpenAI also said the suspected campaign occurred between roughly late 2025 and earlier this year. The operation tried to latch onto news coverage about data centers, then it used ChatGPT to generate critical posts. That linkage is the new wrinkle: OpenAI says that, unlike previous suspected Chinese influence operations it has identified, this one specifically used the company’s AI models to try to sway US opinion on data centers. Nimmo called it “particularly ironic” that the attackers used American AI to do it.

OpenAI stressed that the campaign’s intent was to exploit and amplify public concerns about energy prices and local impacts of data center development. Those topics are not speculative. Data centers do raise real questions for communities, including electricity costs and local effects. That is exactly what influence operators hunt for, because domestic debates already exist and audiences are primed to engage. OpenAI said it did not find evidence of a meaningful breakout beyond its own posts, and Nimmo said they saw no signs of success, but the fact that operators tried at all tells boards and leadership teams something uncomfortable: the narrative testing and automation can be faster than traditional defenses.

OpenAI also detailed how the campaign likely accessed ChatGPT. It said users “prompted ChatGPT in Simplified Chinese while repeatedly asking for English- and Chinese-language outputs.” The company noted that it does not allow access to ChatGPT in China, meaning the campaign needed to rely on VPNs to access the AI tools. That operational detail is important because it points to the enforcement gap rather than just the content. It is one thing to moderate prompts; it is another to assume the attacker can reach the model using workarounds. For executives, the subtext is that security and integrity systems have to anticipate not only what gets generated, but how the generator gets access.

OpenAI’s report also adds a second suspected operation, which it dubbed the “Tech and Tariffs” Campaign. In that one, OpenAI said users used ChatGPT to “generate short comments and political cartoons criticizing US tech policies and tariffs.” OpenAI said some posts included AI-generated cartoon images of President Donald Trump depicting him as careless, including one showing him sawing a ladder while standing on it. OpenAI said the timing was notable because it happened around October 2025, when Trump announced an additional 100% tariff on Chinese goods. Unlike the data center operation, OpenAI said it was unable to link the tariff and political campaign to a particular entity in China. OpenAI did note that the users for this suspected campaign asked that China’s leader, Xi Jinping, be excluded from any AI-generated cartoon images.

The broader consequence for the C-suite is that the data center debate is now more tightly coupled with AI-driven information operations. OpenAI frames the infrastructure stakes as part of the AI race, and it positions data centers as essential for capacity. That means any attempt to sour sentiment, even if it stays “small scale,” can still shift the risk profile for regulators, local governments, and major customers in the ecosystem. If energy and local impacts are the narrative targets, then the downstream effects could include permitting friction, reputational pressure, or investor uncertainty around timelines.

OpenAI’s findings also land in a political moment where claims about data center opponents and potential foreign ties are already being traded publicly. The source notes that in a claim unrelated to OpenAI’s report, “Shark Tank” star Kevin O’Leary suggested that some opponents to his company’s Utah data center critics had ties to the Chinese Communist Party. Two political strategists O’Leary called out by name strongly disputed the claim in a mocking online video. Amid wider pushback to O’Leary’s project and requests from Utah lawmakers, he agreed to cut the size of his proposed data center in half. That sequence is not proof of anything about OpenAI’s suspected campaign. But it shows how quickly infrastructure debates can amplify, politicize, and translate into concrete changes.

For peer leaders running AI products, platform ecosystems, or infrastructure-adjacent businesses, OpenAI’s message is clear: the toolchain that powers legitimate AI can also be routed into influence operations. Nimmo said the suspected campaign is important because it shows the “intentions of influence operators from China, and the narratives they’re testing,” regardless of effectiveness. In other words, even if the posts did not meaningfully break out, the playbook still exists. And when the playbook improves, executives will have less time than ever to protect trust.