OpenAI disclosed on June 10, 2026 that it disrupted two distinct influence operations originating from China that attempted to use ChatGPT to manipulate public discourse in the United States. The first campaign, named ‘Data Center Bandwagon,’ generated social media posts and images claiming that AI data center expansion was driving up electricity costs for American families. The second operation, dubbed ‘Tech and Tariffs,’ used the AI tool to create political cartoons and social media comments criticizing US technology policies and tariffs, with content specifically designed to exclude Chinese President Xi Jinping while focusing attacks on President Donald Trump.

The campaigns operated between late 2025 and early 2026, relying on VPNs to bypass China’s restrictions on ChatGPT access. Users prompted the AI system in simplified Chinese while requesting both English and Chinese-language outputs. They then deployed this content across platforms including X, Facebook, and YouTube while posing as American users. Despite the technical sophistication of using AI for content generation at scale, both operations achieved minimal authentic engagement and failed to meaningfully influence their target audiences.

Ben Nimmo, principal investigator at OpenAI, characterized the activity as a classic foreign influence operation attempting to hijack genuine domestic debates. The timing proved particularly notable given OpenAI’s own financial interests in US data center development, though the company stressed it found no evidence that anti-data center sentiment in America was being driven by foreign propaganda. The revelation comes amid broader concerns about AI governance challenges facing technology companies and governments worldwide.

Private Chinese Technology Firm Traced to Data Center Campaign

OpenAI’s investigation traced the ‘Data Center Bandwagon’ operation to an unnamed Chinese technology company holding multiple contracts with regional Chinese governments. The operators worked as part of a social media operations team conducting influence work for Chinese provincial-level government clients.

The campaign exploited existing public anxieties about energy prices and the local environmental impact of large-scale data center development. Content generated through ChatGPT alleged that data center buildouts were raising electricity costs for ordinary American families, attempting to amplify genuine concerns into coordinated opposition.

However, the operation’s actual impact remained negligible. OpenAI rated the campaign a 1 on the Bookings breakout scale, indicating activity on one or more platforms but no evidence of meaningful engagement by targeted audiences. Most posts gained little to no authentic interaction outside the operators’ own amplification networks.

The perpetrators fed work reports into ChatGPT that inadvertently exposed operational security details about their social media campaigns. These documents described goals of “establishing persistent and credible accounts, producing visually appealing content to expand audience reach in different regions and maintaining long term account viability by anticipating platform enforcement.” Another report analyzed how to leverage Facebook’s content ecosystem, groups, pages, hashtags, advertising tools, and recommendation systems while evading Meta’s detection of coordinated inauthentic accounts.

Nimmo acknowledged the irony of foreign actors using American AI technology to attack US data center development. The circumstances proved especially notable given OpenAI’s own efforts to raise hundreds of millions of dollars in funding to build data centers across multiple locations.

Tariff Campaign Excluded Xi Jinping from AI-Generated Content

The second operation focused on US technology policies and tariffs, using ChatGPT to generate short comments and political cartoons in multiple languages including English, Italian, Japanese, and traditional Chinese. The content accused the United States of prioritizing profits over loyalty to its allies.

Timing analysis revealed the campaign intensified around October 2025, coinciding with President Trump’s announcement of an additional 100% tariff on Chinese goods. The operators specifically directed ChatGPT to only include Trump in generated content while excluding Xi Jinping, despite China’s own extensive use of tariffs as economic policy tools.

The campaign produced AI-generated cartoon images depicting Trump as careless or incompetent, including one showing him sawing a ladder while standing on it. Unlike the data center operation, OpenAI could not link this activity to a specific entity in China, though patterns and tactics overlapped with pre-established Chinese government propaganda campaigns online.

OpenAI rated this campaign slightly higher at a 2 on the Bookings breakout scale, indicating activity across multiple platforms but still no evidence of meaningful authentic engagement. The same network targeted X users with an influence campaign alleging a widespread user data breach that Nimmo confirmed ‘never happened.’

The challenges facing AI deployment extend beyond influence operations, as evidenced by recent findings about confused corporate AI strategies creating internal confusion and wasting resources.

AI-Generated Content Fails to Achieve Viral Traction

Despite the ability of AI tools to create internet content at scale, the Chinese operations consistently failed to generate authentic engagement or viral spread. Images used by the operators appeared clunky or employed overly direct messaging that displayed unfamiliarity with both the English language and internet virality.

Researchers studying state-sponsored influence campaigns noted that foreign groups regularly attempt to piggyback off established narratives with organic momentum. By latching onto genuine domestic movements like public concerns about AI infrastructure and data centers, these operations can appear more effective than their actual impact suggests.

Nimmo emphasized the distinction between creating a debate and attempting to interfere in an existing one. The debate about data centers and their impact on American communities existed independently of foreign interference. Many participants on both sides held reasonable and sincerely held views. The concerning element involved covert foreign actors posing as Americans to shape the conversation, particularly using the very AI technology they attacked in their messaging.

The limited success of these campaigns points to ongoing challenges foreign actors face in achieving authentic influence despite access to advanced AI content generation tools. Third-party engagement remains an imperfect but important indicator of an influence operation’s true impact, and both Chinese campaigns fell far short on this metric.

While OpenAI’s threat intelligence team successfully identified and disrupted these operations, Nimmo stressed the value in understanding the intentions of influence operators from China and the narratives they continue to test, regardless of current effectiveness.

Frequently Asked Questions

How did OpenAI detect these Chinese influence operations?

OpenAI’s threat intelligence team identified two distinct clusters of coordinated activity through behavioral patterns including users prompting ChatGPT in simplified Chinese while requesting English and Chinese outputs, use of VPNs to bypass China’s ChatGPT restrictions, and work reports fed into the system that revealed operational security details about social media campaigns. The team traced one operation to a Chinese technology company with government contracts based on these technical indicators and content analysis.

What impact did these influence campaigns actually achieve?

Both campaigns achieved minimal authentic engagement and failed to meaningfully influence their target audiences. OpenAI rated them 1 and 2 on the Bookings breakout scale, indicating platform activity but no evidence of genuine engagement from targeted users. Most posts gained little to no interaction outside the operators’ own amplification networks. The company found no evidence that anti-data center sentiment in America was being driven or bolstered by these foreign propaganda efforts.

Does OpenAI allow ChatGPT access in China?

OpenAI does not permit access to ChatGPT in China, which forced the influence campaign operators to rely on VPNs to circumvent geographic restrictions. The users prompted the AI system in simplified Chinese while requesting outputs in multiple languages, then deployed the generated content on Western social media platforms like X, Facebook, and YouTube while posing as American users. This violation of terms of service led OpenAI to ban the accounts involved in both operations.

Broader Implications for AI Security and Public Discourse

The disclosure of these influence operations highlights the dual-use nature of AI language models as both content creation tools and potential vectors for information manipulation. While ChatGPT and similar systems enable legitimate applications across countless domains, they also lower technical barriers for foreign actors attempting to generate persuasive content at scale.

The failure of these particular campaigns to gain traction provides limited reassurance. Future operations may improve their cultural fluency and messaging sophistication, making detection and resistance more challenging. The ease with which operators can iterate and test different narratives using AI tools creates a concerning asymmetry favoring attackers over defenders.

OpenAI’s position as both investigator and potential beneficiary of data center development creates complex dynamics. While the company maintained transparency about its financial interests and stressed the authenticity of domestic data center debates, questions about conflicts of interest in threat intelligence reporting merit ongoing scrutiny. Independent verification and cross-referencing with other security researchers becomes essential when threat intelligence comes from parties with direct stakes in the narratives being targeted.

The incidents also illuminate tensions in US-China technology competition. Data centers represent critical infrastructure for AI development, and the focus of Chinese influence operations on undermining support for American data center expansion aligns with strategic economic interests. Similar dynamics appear in debates about semiconductor manufacturing, cloud computing infrastructure, and other foundational technologies.

Detection capabilities demonstrated by OpenAI and other platform operators represent important defenses, but reactive identification of influence operations after deployment provides incomplete protection. The work reports inadvertently fed into ChatGPT revealed sophisticated understanding of platform enforcement mechanisms and strategies for maintaining account viability, suggesting operators actively adapt to defensive measures.

The revelation comes as concerns about AI’s broader societal impacts continue to generate intense public debate. Foreign actors attempting to exploit and amplify these genuine concerns for strategic advantage adds another layer of complexity to already fraught policy discussions about AI governance and infrastructure development.

Conclusion

The Chinese influence operations identified by OpenAI demonstrate both the growing accessibility of AI-powered disinformation tools and the continued challenges foreign actors face in achieving authentic engagement with American audiences. While ChatGPT enabled rapid content generation at scale, cultural and linguistic barriers prevented the campaigns from gaining meaningful traction beyond their own amplification networks.

The specific targeting of data center infrastructure and technology policy debates reveals strategic priorities in US-China competition over AI dominance. As both nations invest heavily in the computational infrastructure required for advanced AI development, information operations attempting to undermine public support for American technology buildouts represent predictable elements of broader rivalry.

Looking ahead, the sophistication of AI-generated influence content will likely improve as operators refine their techniques and develop better understanding of target audience sensibilities. Platform operators, government agencies, and civil society organizations must continue developing detection capabilities and public awareness while grappling with the challenge of distinguishing between foreign manipulation and legitimate domestic debate. The success of defensive efforts will depend on technical detection systems, critical media literacy, and transparent disclosure of both threats and conflicts of interest in threat reporting.