China Used American AI to Fight America's AI Agenda

.

When the Enemy Uses Your Own Weapons

There is something almost cinematic about it: Chinese propagandists, working to undermine American technological leadership, turned to America's own most famous AI tool to do the job. They used ChatGPT. And they got caught.

On June 10, 2026, OpenAI published a threat intelligence report detailing how two distinct clusters of China-linked accounts — both banned from the platform — exploited its chatbot to manufacture social media content designed to stoke American public anger. One campaign targeted the politically sensitive issue of AI data centers and energy costs. The other went after President Trump's trade tariffs and the broader U.S. push for global tech dominance.

Both groups of accounts prompted ChatGPT in Simplified Chinese and used VPNs to access the platform, since OpenAI does not permit access from within China.

Campaign One: "Data Center Bandwagon"

The first cluster, which OpenAI named the "Data Center Bandwagon" campaign, generated English- and Chinese-language social media comments and images claiming that the rapid build-out of AI data centers was driving up electricity costs for ordinary American families.

The operators did not invent their narrative from thin air. In 2025, local opposition blocked or delayed dozens of U.S. data center projects representing more than $150 billion in potential investment. The campaign latched onto that existing anxiety and attempted to sharpen it.

The accounts asked ChatGPT to generate comic strips based on reporting from a legitimate regional newspaper about a power grid operator's capacity auction prices, framing rising electricity demand as a consequence of AI infrastructure — and presenting those costs as ultimately passed on to ordinary households.

The resulting content depicted executives and robots carrying bags of money while "ordinary people" bore the costs of the AI industry. The posts appeared on X, linked to real news articles to lend them an air of credibility.

Targeting Chinese Dissidents

The same campaign had a second, darker dimension. The operators also asked ChatGPT to generate short comments targeting the X account of Chinese dissident Li Ying, known online as "Teacher Li," and other Chinese-language commentators. OpenAI's models refused to generate the inflammatory personal attacks the operators requested.

One particularly notable tactic involved the cluster posing as U.S.-based Chinese immigrants — workers, students, mothers, clerks and investors — to encourage a Chinese-American YouTuber to produce content highlighting America's "dark side." In other words: fabricated personas, manufactured credibility, and real people being manipulated as unwitting mouthpieces.

Campaign Two: "Tech and Tariffs"

The second campaign, which OpenAI dubbed "Tech and Tariffs," focused on generating political cartoons and short comments criticizing U.S. tariffs as an attempt by Washington to seize dominance over global technology markets.

In one cartoon, Trump was depicted wearing American flag pants labeled "America First" while swinging a mallet marked "Tech Dominance" at a wall reading "Global Future." The operators were careful to keep Beijing entirely out of frame: their prompts explicitly instructed ChatGPT not to include Chinese President Xi Jinping in any output — only Trump was to appear.

The operation had a global reach. Beyond English, the campaign generated content in Italian, Japanese and Traditional Chinese, including material targeting Taiwanese audiences and accusing the U.S. of putting profits over loyalty to its allies.

A Side Operation: Smearing ChatGPT Itself

A separate but likely connected network of fake X accounts spread false claims that ChatGPT user data had been compromised and that users' lives had been negatively impacted. OpenAI stated that these allegations were entirely false. The goal appeared to be reputational: discredit the very tool being used against the campaign's operators.

The operators also asked ChatGPT to help design an AI-powered surveillance system capable of automatically scraping information from social media accounts, storing logs and downloading videos for large-scale analysis. OpenAI's models declined to provide the surveillance-specific components of this request.

Did It Work? Mostly No — But That's Not the Point

OpenAI said the influence operations failed to gain significant online traction. Ben Nimmo, principal investigator on OpenAI's intelligence and investigations team, was direct in his assessment: "This was not a case of an influence operation creating a debate. The debate existed already. This was an influence operation from China trying to interfere in it."

Using the industry-standard Breakout Scale, OpenAI assessed both operations as Category One — activity confined to a single platform, with no evidence of broader public amplification.

Yet the significance of the operation lies not in its impact but in its intent. OpenAI drew a direct parallel to earlier PRC-linked operations that targeted rare earth mining companies in the United States, Canada and Australia — companies that Beijing viewed as competitors to China's global dominance in that sector. In both cases, inauthentic accounts targeted private companies in democracies operating in sectors Beijing considered strategically important.

The timing reinforces the pattern. The campaign followed the CCP's Fourth Plenary Session, at which the Party elevated AI as a top strategic priority in its 15th Five-Year Plan and announced a nationwide "AI+" initiative — mirroring how the 14th Five-Year Plan had previously securitized rare earth materials.

The Irony OpenAI Doesn't Let Slide

OpenAI's report closes with a pointed observation: it remains ironic that both operations chose to use American AI — rather than Chinese models — to generate their content attacking American AI. The company said it was not in a position to determine what drove that choice, but noted that China's official cyber strategy actually emphasizes the use of domestically developed open-weights models.

The Chinese Embassy in Washington offered a generic denial, stating that Beijing firmly opposed "groundless attacks or smears" and was committed to ensuring AI serves as "a force for good." It did not address any specifics of the report.

What This Tells Us

OpenAI's findings fit a well-established playbook. China's influence operations — whether targeting rare earths companies, Taiwanese video games, or now American AI infrastructure — do not typically try to manufacture new controversies. They find existing fault lines and press on them.

What is new is the use of generative AI to do it at scale, in multiple languages, with operationally sophisticated planning documents uploaded directly into ChatGPT. The tool was turned against its own industry — and against the broader U.S. technology agenda it represents.

OpenAI concluded that both operations targeted narratives likely to remain attractive for Chinese influence efforts precisely because they can be slipped into legitimate American debates, nudging audiences toward distrust of U.S. institutions, technology companies and democratic policy choices — all while advancing Beijing's strategic position in AI development.

.

Sources

1. OpenAI – June 2026 Threat Report (full PDF): https://cdn.openai.com/pdf/96b559fa-c165-4575-805d-e636909e2f78/June-2026-Threat-Report.pdf
2. OpenAI – Official announcement page: https://openai.com/index/prc-linked-influence-operations-ai-debates/
3. Reuters (via KFGO): https://kfgo.com/2026/06/10/openai-says-chinese-propaganda-is-being-deployed-to-foment-dissent-over-tariffs-data-centers/
4. Axios – China-based operatives used ChatGPT: https://www.axios.com/2026/06/10/openai-china-ai-data-center-tariffs-chatgpt
5. CyberScoop – OpenAI China influence campaign: https://cyberscoop.com/openai-china-influence-campaign-chatgpt/
6. The Next Web – OpenAI China data center influence: https://thenextweb.com/news/openai-china-data-center-influence-campaign

.