X Exposes 200,000-Account China-Linked Bot Network

A new disclosure from X has added to the growing body of evidence that China-linked actors are using large networks of inauthentic social-media accounts to influence political, economic and technological debates outside China. X says its safety teams identified a network of approximately 200,000 suspected Chinese inauthentic accounts, including a smaller group that was actively attempting to shape discussion in the United States around artificial-intelligence data centres, electricity prices and pressure on power grids.

The findings fit into a much wider pattern already documented by Google, Meta, Microsoft, OpenAI and US authorities. In several of those earlier cases, investigators were able to connect influence operations more directly to Chinese state-linked organisations or individuals associated with Chinese law enforcement.

X Identifies a Large Inauthentic Network

X’s Global Government Affairs operation disclosed in late August 2026 that the platform had investigated a suspected Chinese bot farm involving roughly 200,000 accounts. Within this much larger network, around 200 accounts were identified as posting material intended to influence American debate over AI infrastructure and energy policy.

The content focused heavily on claims that the expansion of large AI data centres was increasing household electricity costs, placing additional pressure on regional power grids and benefiting major technology companies at the expense of ordinary consumers. Some of the accounts circulated cartoons and other visual material portraying data-centre operators negatively, while others amplified legitimate news reports and statistics about electricity demand in ways designed to reinforce a particular narrative.

This is an important characteristic of contemporary influence operations. They do not necessarily depend on fabricated information. A coordinated network can be more effective when it attaches itself to an existing controversy, selectively amplifies genuine concerns and attempts to make one interpretation of the issue appear more widespread than it actually is.

OpenAI Had Already Detected a Related China-Origin Campaign

The X discovery becomes more significant because OpenAI had already identified a similar operation earlier in 2026. In June, OpenAI disclosed a campaign it called “Data Center Bandwagon,” involving accounts assessed as likely originating in China that used generative AI to create English-language posts and images criticising American data-centre expansion.

The operators primarily interacted with AI systems in Simplified Chinese while requesting material that would appear to have been written by Americans from different backgrounds. The generated content was then distributed through likely inauthentic social-media accounts and focused on themes similar to those later identified by X, including electricity costs, grid pressure and the argument that ordinary consumers were effectively subsidising large technology companies.

OpenAI also found that the operation frequently relied on genuine media reporting about electricity markets and data-centre demand. This further demonstrates how influence campaigns can mix authentic information with coordinated messaging in order to make their narratives appear organic and credible.

Possible Connection to a Chinese Technology Company

OpenAI’s investigation provided a more detailed indication of who may have been involved in the earlier campaign. The company assessed that the operators were likely members of a social-media operations team at a private Chinese technology company carrying out work for provincial-level government clients in China.

That finding is important because modern state-linked influence activity does not necessarily have to be conducted directly by government employees. Governments can use private contractors, technology companies, public-relations firms or other intermediaries to perform online operations, creating organisational distance between state institutions and the people actually running fake accounts.

Such arrangements also complicate attribution. A network may be China-based and working for government clients without every individual account being directly controlled from a central government office. This is one reason why careful terminology such as “China-linked,” “PRC-linked” or “suspected Chinese inauthentic activity” is often more accurate than automatically describing every operation as a direct CCP campaign.

The Same Infrastructure Was Used Against Chinese Dissidents

OpenAI’s findings also showed that the Data Center Bandwagon operation was not limited to US infrastructure policy. Operators associated with the campaign requested material attacking Li Ying, a prominent Chinese dissident and online commentator known as “Teacher Li.”

The overlap between outward-facing influence activity and attacks on Chinese dissidents is particularly significant. It suggests that the same networks, operators or technical infrastructure can potentially serve multiple objectives, including shaping foreign political narratives, discrediting critics of the Chinese government and monitoring online reactions.

This pattern has appeared in other investigations into China-linked information operations. Influence campaigns and transnational repression can therefore overlap, with fake accounts serving both geopolitical and domestic political purposes.

Google Has Tracked DRAGONBRIDGE for Years

Google has documented one of the largest known China-linked influence ecosystems through its investigations into DRAGONBRIDGE, also known as Spamouflage Dragon. Google describes the network as a prolific influence operation linked to the People’s Republic of China and has removed enormous numbers of associated channels and accounts over several years.

By early 2024, Google had disrupted more than 175,000 instances of DRAGONBRIDGE activity across YouTube and Blogger. The network produced material covering Taiwan, US politics, international conflicts and other geopolitical issues, while some accounts attempted to present themselves as residents of foreign countries or reproduce content from legitimate users.

Google’s continuing removals during 2026 show that the operation has not disappeared. China-linked coordinated influence networks continue to re-emerge across platforms, often using new accounts and modified tactics after earlier infrastructure is removed.

Large Networks Do Not Automatically Mean Large Influence

Google’s research also provides an important warning against equating scale with effectiveness. Although DRAGONBRIDGE generated enormous quantities of material, much of it attracted very little genuine engagement.

In 2023, Google found that the majority of tens of thousands of DRAGONBRIDGE YouTube channels it removed had no meaningful subscriber base, while large numbers of uploaded videos received fewer than 100 views or no views at all. Much of the visible interaction appeared to come from other accounts within the same coordinated ecosystem rather than from genuine users.

Meta Linked Spamouflage Activity to Chinese Law-Enforcement Personnel

Meta’s investigations have produced some of the strongest public attribution of China-linked influence operations. In 2023, the company announced the removal of what it described as the largest known cross-platform covert influence operation in the world, involving activity across more than 50 online services.

Meta said it had been able to connect elements of the operation, widely known as Spamouflage, to individuals associated with Chinese law enforcement. The network operated not only on Facebook and Instagram but also across X, YouTube, TikTok, Reddit and numerous other websites and social platforms.

This attribution is stronger than simply determining that accounts originated in China. It demonstrates that at least part of the broader influence ecosystem had direct links to individuals associated with Chinese state institutions.

Earlier Meta Investigations Found Similar Patterns

Meta had identified related activity before the large 2023 takedown. In 2019, the company removed accounts involved in coordinated inauthentic behaviour targeting the Hong Kong protests and said its investigation found links to individuals associated with the Chinese government.

In 2022, Meta also dismantled a China-origin network targeting audiences in the United States and Europe. Rather than consistently supporting one political position, some of the accounts engaged with people on opposing sides of controversial issues, suggesting that the objective was sometimes to intensify division rather than persuade users to support a specific Chinese policy.

This tactic is particularly important because it changes the traditional understanding of propaganda. An influence operation does not necessarily need to make foreign audiences admire the country behind it. Creating greater distrust, confusion and polarisation within another society can itself produce strategic value.

Microsoft Found CCP-Affiliated Accounts Posing as Americans

Microsoft has independently documented similar activity through its Threat Analysis Center. In 2024, the company reported that CCP-affiliated actors were using deceptive social-media accounts to pose as American voters and engage with politically sensitive issues in the United States.

Some of the accounts asked users questions about controversial domestic subjects, apparently testing which themes generated the strongest response. Microsoft also observed increasing experimentation with AI-generated images and political material designed to appear as though it had been created organically by American users.

The findings suggested a shift away from obvious pro-China messaging towards more sophisticated efforts to participate in local political conversations while disguising the foreign origin of the participants.

Generative AI Changes the Economics of Influence Operations

Artificial intelligence is making these campaigns faster, cheaper and easier to scale. Historically, operating convincing social-media personas in another country required large teams of people who understood local language, political culture and social conventions.

Generative AI can automate a significant portion of that work. A small group of operators can produce hundreds of posts, translate material into multiple languages, imitate different writing styles, generate political cartoons and continually modify messages according to audience response.

The result is not necessarily more persuasive propaganda, but it allows influence networks to experiment at a much larger scale. Thousands of messages can be generated and tested, while only the narratives that gain traction need to be amplified further.

Real Social Divisions Provide the Best Targets

The most effective influence campaigns frequently focus on issues that already divide societies. Energy prices, immigration, environmental policy, elections, racial tensions, trade restrictions, military conflicts and public-health controversies all provide opportunities because genuine disagreements already exist.

The operator does not have to create the dispute. Instead, coordinated accounts can amplify the most divisive arguments, promote extreme interpretations and make public disagreement appear larger or more hostile than it actually is.

The AI data-centre controversy fits this model particularly well. Concerns about electricity consumption and infrastructure are real, but a foreign influence network can exploit those concerns by posing as local citizens and artificially magnifying one side of the debate.

US Authorities Have Alleged Direct State Use of Fake Personas

Evidence of Chinese state involvement in social-media manipulation also exists outside the technology companies. In 2023, the US Department of Justice charged 34 officers of China’s Ministry of Public Security over an alleged operation involving thousands of fake online identities.

According to prosecutors, the officers were members of a unit known as the 912 Special Project Working Group and allegedly used false social-media personas to promote official Chinese government narratives, attack critics and influence perceptions of China and the Communist Party.

Some of the accounts allegedly posed as Americans, creating the appearance that US citizens independently supported positions favoured by Beijing. These remain criminal allegations rather than convictions, but the case provides unusually detailed public evidence of how Chinese state personnel are alleged to have organised covert online influence activity.

Propaganda and Transnational Repression Can Use the Same Networks

The Justice Department allegations also demonstrate how propaganda operations can overlap with transnational repression. The same infrastructure that allegedly promoted favourable narratives was also used to harass critics and dissidents living outside China.

OpenAI’s 2026 findings show a similar pattern, with operators involved in the data-centre campaign also generating material targeting a Chinese dissident. This suggests that fake-account networks may be built as flexible infrastructure capable of supporting several political objectives simultaneously.

Such networks can therefore be used for foreign influence, political harassment, narrative management and intelligence gathering at the same time.

Attribution Remains Difficult

Identifying fake behaviour is generally easier than determining who ultimately controls it. Operators can use VPNs, proxy servers, stolen accounts, foreign hosting infrastructure and contractors to obscure their location and organisational links.

This is why major technology companies use different levels of attribution. Google describes DRAGONBRIDGE as PRC-linked, Microsoft has identified CCP-affiliated actors in specific operations, Meta has tied Spamouflage activity to individuals associated with Chinese law enforcement, and OpenAI assessed certain operators as belonging to a private Chinese company serving provincial government clients.

X has so far used more cautious terminology for the latest network. Until stronger evidence is published, the new 200,000-account operation is best described as suspected Chinese inauthentic activity or a China-linked bot farm, rather than as a definitively proven CCP-run network.

What the X Discovery Establishes

The most defensible conclusion from the August 2026 disclosure is that X uncovered a very large network of suspected Chinese inauthentic accounts and identified a smaller group actively attempting to influence debate over US AI infrastructure and energy policy.

The discovery is consistent with the earlier Data Center Bandwagon operation identified by OpenAI and with the broader history of China-linked influence activity documented by Google, Meta and Microsoft.

An Expanding Global Contest Over Information

The wider significance of the episode extends well beyond China and the United States. Social-media influence operations have become a permanent component of geopolitical competition, giving governments, contractors, intelligence organisations and ideological groups inexpensive tools for reaching foreign populations directly.

The barrier between domestic political debate and international strategic competition has consequently weakened. A discussion that appears to involve ordinary residents arguing about electricity prices or infrastructure can potentially include automated or coordinated identities operated from another country.

The same techniques can be applied to elections, military conflicts, trade disputes, social tensions, energy policy or public-health debates. The purpose may be to promote a specific position, but it can also be to increase distrust over which voices online are authentic.

The Challenge for India

The implications are highly relevant for India, which has one of the world’s largest and most politically active online populations. Public debate takes place across numerous languages and platforms, creating a vast information environment that foreign influence operators could attempt to exploit.

Sensitive issues involving border disputes, defence policy, communal tensions, elections, international relations or major infrastructure projects could all become potential targets. Generative AI makes this easier because foreign operators no longer require large teams fluent in Hindi, Bengali, Tamil or other Indian languages to produce locally styled content at scale.

India’s information-security challenge will therefore increasingly extend beyond conventional cyberattacks. Protecting networks and databases remains essential, but identifying attempts to manipulate citizens through synthetic identities and coordinated narratives will become equally important.

Synthetic Consensus Can Be as Important as Viral Propaganda

Bot farms can influence perception even when individual posts do not become viral. Thousands of apparently independent accounts repeating similar messages can create the impression that a particular opinion enjoys overwhelming public support.

This phenomenon can be understood as synthetic consensus. A user encountering hundreds of accounts making the same argument may reasonably assume that they are witnessing an authentic grassroots reaction, even when those accounts belong to a coordinated operation.

Such networks can also attempt to influence recommendation systems, trending topics and search visibility. Their objective may therefore be less about persuading every user directly and more about manipulating the information environment in which genuine users form their opinions.

Platforms Face a Difficult Detection Problem

Social-media companies face the difficult task of identifying coordinated manipulation without suppressing legitimate political expression. Users who genuinely hold the same opinion will naturally post similar material, and a platform cannot treat agreement itself as evidence of foreign interference.

Detection therefore depends on behavioural indicators such as synchronised posting, shared technical infrastructure, unusual account-creation patterns, repeated content, linked login behaviour and hidden relationships between apparently independent profiles.

Influence operators continuously adapt to those detection systems. They can allow accounts to age before activating them, mix political posts with ordinary personal content, vary posting times and use generative AI to make each persona appear more individual.

Cross-Platform Cooperation Is Becoming Essential

No single technology company can observe the entire lifecycle of a modern influence operation. Content may be generated using one AI platform, coordinated through a messaging service, published on one social network and amplified across several others.

Meta’s Spamouflage investigation demonstrated this clearly by identifying activity across more than 50 platforms and online services. This means effective detection increasingly requires information sharing between social-media companies, AI developers, cybersecurity researchers and government agencies.

A suspicious behavioural pattern identified by one platform may help another discover related accounts. Similarly, an AI company detecting organised content generation can provide early warning before that material is distributed widely elsewhere.

A New Phase of Information Competition

X’s discovery of approximately 200,000 suspected Chinese inauthentic accounts illustrates how information competition is becoming increasingly automated and difficult to detect. Influence operations that once required large propaganda organisations can now be conducted through thousands of digital identities that appear to be ordinary individuals.

Generative AI can write their posts, create their images and help adapt their language to particular audiences. Automated systems can schedule activity continuously, while analytics can identify which narratives generate the greatest response.

The subject of the latest campaign is itself revealing. Rather than focusing solely on traditional geopolitical propaganda, the accounts entered a debate over AI data centres, electricity infrastructure and the economic consequences of artificial intelligence. These are technologies that are likely to shape national power and industrial competitiveness for decades.

The competition surrounding AI is therefore no longer limited to chips, algorithms and computing infrastructure. It increasingly extends into the political environment in which those technologies are developed.

X’s latest takedown does not establish that a 200,000-account Chinese network successfully altered American opinion, nor does it prove direct CCP control of every account involved. What it does demonstrate, when combined with findings from OpenAI, Google, Meta, Microsoft and US authorities, is a sustained pattern in which China-linked actors have experimented with fake personas, coordinated account networks, artificial intelligence and existing social divisions in attempts to influence audiences outside China.

The challenge for democratic societies will increasingly be not only to determine whether information is true or false, but also to understand who is creating the apparent public conversation, whether the participants are genuine, and whose interests may lie behind voices that appear spontaneous but are in fact manufactured.

---