Was It a Mistake? How ChatGPT Sparked a Debate on China’s Global Power
A Chinese law enforcement official used OpenAI’s ChatGPT as a personal diary — and in doing so, pulled back the curtain on one of the most sophisticated transnational repression campaigns ever documented.
It began with a single ChatGPT account. Somewhere within China’s law enforcement apparatus, an official sat down with the world’s most popular AI chatbot — not to build a weapon, not to write propaganda, but to keep notes. To log the day’s work. To think out loud about operations unfolding across continents.
That habit of treating an AI tool like a digital diary would prove to be one of the most consequential security lapses in recent intelligence history. When OpenAI published its latest threat report on February 25, 2026, the details it contained offered a rare, unfiltered glimpse into what modern authoritarian repression actually looks like — from the inside out.
- Hundreds of Chinese operators allegedly involved in the campaign
- Thousands of fake social media accounts used to silence critics
- Targets included dissidents in the United States, Japan, and beyond
- Tactics ranged from forged court documents to fabricated obituaries
- OpenAI banned the account after discovering the activity
The Diary That Wasn’t Private
The unnamed Chinese law enforcement official had been using ChatGPT not to generate content for the operation itself — other AI tools were handling that — but to organize and reflect on it. According to OpenAI’s investigators, the account functioned as a running log of what the user called “cyber special operations”: a structured, state-backed campaign to locate, monitor, and silence Chinese nationals who had dared to criticize the Communist Party from abroad.
The breadth of what was recorded is staggering. The user documented impersonation campaigns, describing how Chinese operators disguised themselves as U.S. immigration officials and contacted dissidents living in America, warning them that their public statements had supposedly violated the law. The goal, investigators believe, was to generate fear — to make the target feel watched, exposed, and legally vulnerable, even on foreign soil.
“It’s not just digital. It’s not just about trolling. It’s industrialized. It’s about trying to hit critics of the CCP with everything, everywhere, all at once.”— Ben Nimmo, Principal Investigator, OpenAI
In another documented case, the operator described efforts to use forged documents — fabricated to look like official paperwork from a U.S. county court — in an attempt to force a social media platform to remove a dissident’s account. The forgery was a calculated gamble on platform compliance and bureaucratic complexity: if the document looked official enough, perhaps no one would look too closely.
An Industrial Operation, Not a Lone Actor
What makes this case so significant is not one tactic or one target, but the scale and architecture of what was uncovered. OpenAI’s investigators found evidence pointing to hundreds of operators and thousands of fake online accounts spanning multiple platforms — a coordinated infrastructure designed not just to harass, but to overwhelm.
Targets were apparently chosen for their visibility and influence within diaspora communities. The campaign did not merely aim to silence individuals; it sought to make examples of them, to signal to other potential critics what dissent might cost. In one documented instance, operatives reportedly filed thousands of automated complaints against a prominent activist’s posts on X, simultaneously flooding the platform’s moderation queue while creating dozens of fake profiles using the activist’s own likeness — a tactic designed to generate confusion and erode credibility.
In perhaps the most chilling detail, the ChatGPT user described the creation of fake obituaries and gravestone photographs for a living dissident, designed to spread false reports of the person’s death across social media. According to OpenAI, false rumors matching this exact description surfaced online in 2023 — providing investigators with a rare opportunity to match the diary’s confessions to real-world events.
Tactic 01
Impersonating U.S. immigration officials to threaten dissidents with fabricated legal jeopardy.
Tactic 02
Using forged U.S. court documents to demand social media platforms remove dissident accounts.
Tactic 03
Filing mass complaints and creating impersonator accounts to overwhelm platform moderation.
Tactic 04
Spreading fabricated obituaries and gravestone photos to simulate a dissident’s death.
Tactic 05
Coordinating smear campaigns against foreign political figures who publicly criticize China.
Tactic 06
Using AI tools to plan and organize operations; ChatGPT refused direct assistance but documented intent.
When AI Refused — But the Network Didn’t
One of the most telling episodes in OpenAI’s report involves Japan. After Sanae Takaichi became Japan’s prime minister in late October 2025, the same ChatGPT user attempted to use the platform to plan a coordinated smear campaign against her, apparently in response to her public criticism of China’s human rights record in Inner Mongolia. The proposed operation involved fake foreign residents sending complaints to Japanese politicians, negative comment flooding across platforms, and amplified hashtags designed to erode her public support.
ChatGPT refused. The system declined to assist with the political targeting operation. But the refusal only tells part of the story. Investigators discovered that hashtags matching the user’s described strategy appeared on Japanese online communities in late October anyway — suggesting that other tools, less scrupulous than ChatGPT, stepped in where OpenAI’s model would not go.
This is the uncomfortable truth embedded in OpenAI’s findings: AI safety guardrails are meaningful, but they are not a full solution. A determined state actor with access to multiple AI systems — and with the resources to deploy hundreds of human operators alongside them — can route around a single platform’s refusals. The diary reveals intent; the hashtags reveal execution.
The Bigger Picture: AI in the Authoritarian Toolkit
OpenAI’s report arrives during an especially fraught moment in the global competition over artificial intelligence. The United States and China are locked in a race for AI supremacy that carries implications far beyond Silicon Valley boardrooms — touching military strategy, diplomatic leverage, and the shape of the information environment that billions of people inhabit.
For Michael Horowitz, a former Pentagon official focused on emerging technologies, the report is a data point that fits a disturbing pattern. The operation, he noted, clearly demonstrates how China is actively deploying AI tools to enhance information operations — not just to generate content, but to plan, coordinate, and track campaigns at scale. The ChatGPT diary is, in his framing, a window not onto an anomaly, but onto a system.
Ben Nimmo, OpenAI’s principal investigator, offered what may be the most precise summary of what the report reveals: what was uncovered is not trolling, not mere online harassment. It is industrialized transnational repression — a manufacturing line of psychological pressure, legal intimidation, identity manipulation, and coordinated silencing, run by a state apparatus against its own citizens who have sought safety by crossing a border.
For journalists, researchers, and policymakers, the case raises an uncomfortable question that will only grow more urgent: as AI tools become cheaper, more capable, and more widely available, how many other diaries are out there — still undetected, still running?
Frequently Asked Questions
How did OpenAI discover this operation?
A Chinese law enforcement official used ChatGPT as a personal logbook to document their covert operations. OpenAI’s trust and safety investigators identified the account and its activity during routine monitoring, then matched the documented tactics against real-world events — such as false death rumors that appeared online in 2023 — to verify the account’s confessions corresponded to actual campaigns.
What did ChatGPT actually help with?
ChatGPT served primarily as a journal and organizational tool for the operator, not as a content generator for the operation. When directly asked to help plan a political smear campaign — specifically against Japan’s prime minister — ChatGPT refused. Most of the operation’s content was generated by other AI tools and spread through separate networks of fake accounts.
Who were the targets of this campaign?
The primary targets were Chinese dissidents living abroad — particularly those in the United States — who had publicly criticized the Chinese Communist Party. The campaign also extended to foreign political figures, including Japan’s prime minister, who had spoken out against China’s human rights record. Individual activists were targeted with identity-based attacks designed to discredit or silence them.
How large was this operation?
According to OpenAI’s report, the operation involved hundreds of Chinese operators working through thousands of fake accounts across multiple social media platforms. This makes it one of the largest documented transnational repression campaigns uncovered through AI platform monitoring.
Is this the first time China has been linked to AI-powered influence operations?
No. OpenAI took prior action in October 2025 against suspected Chinese government-linked accounts that had attempted to use ChatGPT to design social media surveillance tools capable of scanning platforms like X, Facebook, Instagram, Reddit, TikTok, and YouTube for political content and “extremist speech.” The February 2026 report represents a significant escalation in the documented sophistication of these efforts.
What happened to the account after OpenAI discovered it?
OpenAI permanently banned the user account after discovering the activity and documenting its contents. The company published its findings in a public threat intelligence report released on February 25, 2026. CNN and other outlets requested comment from the Chinese Embassy in Washington, D.C., but no response was included in initial reporting.
What does this mean for AI safety going forward?
The case illustrates both the value and the limits of AI platform safety measures. ChatGPT’s refusal to assist with specific tasks was meaningful — but the operation continued using other tools. Experts argue this points to the need for coordinated cross-platform monitoring, international policy frameworks for AI-enabled influence operations, and greater transparency from AI companies about detected state-sponsored misuse. Sources: OpenAI Threat Intelligence Report (Feb. 25, 2026) · CNN Politics · Technobezz · KESQ / CNN Wire