OpenAI Report Exposes CCP's New AI Water Army Tactics, Toronto Square Face Leaks Info
OpenAI's threat report released on June 10th peeled back another layer of the CCP's overseas influence operations: it's not just about flooding platforms with 'wumao' accounts or having state media accounts shout slogans, but is now integrating generative AI into troll farms, public opinion monitoring, identity impersonation, shaping overseas Chinese narratives, and platform evasion processes.
OpenAI stated that the company banned two clusters of ChatGPT accounts suspected to be from China. These accounts used simplified Chinese to instruct the model, accessed ChatGPT via VPN, and then requested output in English, Chinese, Italian, Japanese, and Traditional Chinese. The report named the two clusters "Data Center Bandwagon" and "Tech and Tariffs." The former generated public opinion around the construction of AI data centers in the US, claiming that data centers increase electricity demand, leading to higher electricity bills for ordinary households. The latter generated short comments and political cartoons on topics such as US tariffs, technology policy, AI, 5G, rare earths, new energy, and industrial resilience, portraying the US as pursuing technological hegemony and rule-making power.
The most striking point of the report is not the influence of these accounts. OpenAI's own assessment is that most of these operations remained within a single platform, with almost no diffusion among real users, constituting low-level spillover. However, such operations reveal that the CCP's influence operations are testing which narratives, which topics, and which identity packaging can be inserted into the internal debates of democratic societies. The US already has controversies regarding data center power consumption, electricity prices, and local environmental impacts. The CCP's troll farms do not need to invent contradictions; they only need to exploit existing ones, directing the genuine concerns of ordinary people towards distrust of US AI infrastructure, tech companies, and democratic decision-making.
The OpenAI report also disclosed that the operators of the suspected "Data Center Bandwagon" cluster might belong to the social media operations team of a private Chinese tech company and provide services to provincial government clients in China. Related accounts uploaded work reports detailing the establishment of long-term credible accounts, evading platform detection, building credibility through relatable personas, and creating the illusion of natural spread through cross-account interaction. This is no longer a few stupid keyboard warriors taking orders from within the Great Firewall to curse people; it is a commercialized, outsourced, and process-oriented propaganda engineering effort by the Party-state. The CCP is best at outsourcing dirty work to companies, packaging the companies' work as market behavior, and then claiming ignorance or accusing others of smearing China when problems arise.
The part of the report most directly related to the overseas Chinese-speaking community is that these accounts targeted Chinese dissidents and political commentators. OpenAI stated that the accounts queried public information about Li Ying, also known as "Teacher Li Not Your Teacher," and attempted to generate short comments insulting her team's accounts, but the model refused to generate inflammatory or personally attacking content. The report also mentioned that other Chinese political commentators targeted for harassment included Lu Yiheng, Xu Chi, and the X account @SydneyDaddy1. Duolun Duolian mentioned in his retelling that these targets included Teacher Li, Xu Mouren, Duolun Duolian, and Sydney Daddy; according to the original OpenAI report text, the public text explicitly lists Li Ying, Lu Yiheng, Xu Chi, and @SydneyDaddy1, and describes influence attempts on "a former Chinese police officer and YouTuber in the US," but does not directly name this YouTuber in the visible text.
This case of a "former CCP police officer and US YouTuber" is more alarming than ordinary insults. OpenAI stated that the relevant accounts attempted to impersonate Chinese immigrants, workers, students, mothers, shopkeepers, and investors living in the US, sending messages to this YouTuber encouraging him to talk more about US policy failures and expose the "dark side" of America. This is not buying someone off in the traditional sense, nor does it necessarily require direct contact with CCP agents. Its tactic is to make the target believe they are hearing the "real voice of overseas Chinese," and then gradually shift their focus, emotional direction, and judgment of friend or foe. The advanced form of united front work has never been to force you to speak at gunpoint, but to make you feel that those words were your own thoughts all along.
The "Tech and Tariffs" cluster is more like a content factory for anti-US global issues. OpenAI stated that the relevant accounts generated bulk short comments and articles in Chinese, including supporting CCP narratives, attacking the US and Israel, amplifying anti-Semitic rhetoric, and harassing Chinese dissidents. The accounts also requested content targeting Taiwanese users in Traditional Chinese, and generated materials in Japanese and Italian. Taiwan, the US, overseas Chinese, anti-CCP independent media, the AI industry, and the rare earth industry can all be incorporated into the same influence network. The CCP's foreign propaganda has never been a single-track operation; it plays multiple cards simultaneously: nationalism, anti-Americanism, anti-capitalism, anti-Israelism, anti-democratic chaos, and discrimination against Chinese people. Whichever card can be used is pushed to the target audience.
The report also mentioned that the relevant accounts asked ChatGPT to design a public opinion monitoring system concept for capturing so-called "harmful information," tracking "key personnel," saving logs, downloading videos for large-scale semantic analysis, and sending risk alerts. OpenAI stated that the model only provided general data storage and management advice and did not offer solutions for collecting monitoring data. This detail is very Chinese: using US AI to generate anti-US content while simultaneously figuring out how to monitor people more effectively. The Party-state machine's first reaction to technology has never been to create freedom, but to expand control.
After forwarding and interpreting this report, Duolun Duolian pointed the finger at a more insidious influence path: the CCP's troll farms not only curse anti-CCP individuals but also impersonate overseas Chinese, students, and immigrants, using methods like "submissions," "suggestions," "complaints," and "public opinion feedback" to influence bloggers, urging them to oppose the CCP less and the US more, or even gradually shift their main focus from the CCP to the US, the democratic camp in Taiwan, or other anti-CCP individuals. He further criticized Erdaye's recent content for being excessively focused on topics like Trump and the DPP, suggesting that even if not bought off, such outcomes could objectively fall into the CCP's trap.
Erdaye's response provided another piece of the puzzle. He stated that as a media personality, his email has long received organized and large-scale harassment emails, one type of which is the disguised submission mentioned in the OpenAI report. He summarized that these emails have two purposes: one is to impersonate a fan of a big V to curse and provoke, causing media personalities to fight among themselves; the other is to impersonate Chinese citizens to complain about controversial US issues, hoping to be relayed by the media personality. He believes these emails are usually easy to identify due to batch sending and similar writing styles, and he has set up filters, with much content going directly to spam. He also refuted Duolun Duolian's equating criticism of Trump with opposing the US or being manipulated, emphasizing that his criticism of US policies comes from personal judgment.
This debate itself illustrates a problem: the trouble with the CCP's overseas influence operations is not that everyone who criticizes the US, the DPP, or Trump is a CCP troll. Such crude labeling will only destroy judgment. What is truly troublesome is that the CCP uses real controversies, real disagreements, and real emotions as fuel. Of course, the US can be criticized, the DPP can be criticized, Trump can be criticized, and the opportunists, scammers, and traffic farmers within the anti-CCP circle can also be criticized. The problem is that when a Chinese public account consistently loses its main focus on the CCP, steadily shifts its main firepower to internal contradictions within the democratic camp, and becomes increasingly downplaying the CCP's structural tyranny, readers have reason to ask: what political effect does this content line serve?
Duolun Duolian said that his inbox does not contain the kind of emails he described as "impersonating fans to sow discord," but only a large number of curses and disguised submissions. He stated that he and some others are promoting offline, community, institutional, and government-level actions to expose and expel the CCP's agents and state media propaganda overseas. He also mentioned the repatriation of Italian personnel, calling it the result of the efforts of Xu Mouren, Teacher Li, and others. Since this part belongs to the public statements of the individuals involved, specific action details and case chains require further verification with public materials and cannot be presented as facts fully confirmed by a third party.
But the direction is correct. Overseas anti-CCP efforts cannot stop at just cursing the CCP, forwarding news, making programs, or writing posts. The CCP's overseas united front work, harassment, agent activities, fake submissions, fake fans, fake overseas Chinese public opinion, and fake American commentary are precisely the areas where overseas anti-CCP individuals can most easily collect evidence, form public cases, and push platforms and governments to take action. Those who are harassed should preserve email headers, account links, screenshots, timelines, talking points, and evidence of batch sending; those with technical expertise should organize account networks, language patterns, synchronized forwarding, platform registration times, and interaction relationships; those with legal and media knowledge should submit cases to platforms, security research institutions, parliamentary offices, local law enforcement agencies, and the media. This is more effective than empty calls for revolution and more useful than accusing each other on Twitter.
This is also where the complexity of the CCP's influence operations lies. It has overt state media propaganda, united front work through overseas Chinese associations, pressure from consulates, commercial outsourcing, spontaneous support from 'pinkies,' troll farm companies, fake accounts, disguised submissions, harassment of dissidents, exploitation of identity anxieties among overseas Chinese, and leveraging controversies in democratic societies. It doesn't need every
As long as they continue to create noise, sow discord, pollute issues, and drag down trust, they can deplete overseas anti-communist forces. The CCP's favorite scenario is for the anti-communist circle to constantly accuse each other of being spies, call each other liars, and consume each other, while no one compiles evidence into materials that can get platforms to ban accounts, governments to investigate, or agents exposed.
The value of this OpenAI report is not to prove that a specific blogger is a spy, nor to provide new ammunition for infighting within the Chinese-speaking community. What it truly offers is an operational reminder: CCP influence operations have incorporated AI into their daily toolkit, using fake identities, fake public opinion, fake submissions, fake comics, fake comments, and fake public opinion monitoring plans to process real social conflicts into narrative weapons that serve the interests of the party-state. What overseas anti-communist individuals should do is not to shout "paid" at anyone whose views they dislike, but to record the harassment, submissions, contacts, threats, and united front activities they have experienced, compile them into evidence, and hand them over to those who can handle them. Anti-communism is not just a verbal stance; it also includes prying open, one by one, the dirty hands that the CCP extends overseas.