A Chinese Official Used ChatGPT as a Diary, Exposing a Repression Operation

TL;DR: A Chinese law enforcement official used ChatGPT like a personal diary to document his work, and his work was running transnational repression campaigns against Chinese dissidents living abroad. OpenAI's threat researchers found the logs, correlated them with real-world operations, and banned the account. What they discovered: operators impersonating US immigration officials to threaten deportation, forging US court documents to get social media accounts deleted, creating fake obituaries and gravestone photos for living dissidents, and running influence campaigns against Japan's incoming prime minister. One report the official referenced claimed 300+ operators in a single Chinese province running similar operations. "This is what Chinese modern transnational repression looks like," OpenAI's Ben Nimmo said. "It's industrialized."

The Accidental Confession

Most people use ChatGPT to write emails or debug code. This user used it as a work diary.

The Chinese law enforcement official logged his activities: planning influence operations, tracking dissident suppression efforts, documenting tactics used against overseas targets. OpenAI's threat intelligence team, analyzing account activity patterns, found the logs and started connecting dots.

What emerged was a detailed picture of what the Chinese government calls "persuasion to return" and what everyone else calls transnational repression: organized campaigns to harass, intimidate, and silence Chinese dissidents living outside China's borders.

On February 25, 2026, OpenAI published its findings. They banned the account. And for the first time, outsiders got a window into how these operations actually work.

The Playbook

The ChatGPT logs revealed a toolkit of intimidation tactics used against US-based Chinese dissidents:

Impersonating US immigration officials: Operators called dissidents pretending to be ICE or CBP agents. They threatened deportation based on fabricated violations. "Your public statements have broken the law." A lie designed to terrify.
Forging US court documents: When intimidation calls didn't work, operators forged documents that appeared to come from US county courts. They sent these to social media platforms demanding account takedowns for legal violations that didn't exist.
Faking deaths: In at least one case, operators created a fake obituary for a living dissident. They mocked up photos of a gravestone. They posted both on Chinese-language forums to spread the rumor that the target had died. The false death circulated in Chinese media as early as 2023.
Mass harassment campaigns: Thousands of fake accounts across multiple platforms, coordinated to flood targets with attacks and overwhelm their ability to communicate.

This isn't trolling. This is a state apparatus running psychological warfare against its own citizens who happened to leave.

"Industrialized"

How big is this operation? The ChatGPT user referenced a report claiming at least 300 operators in a single Chinese province running similar campaigns. Similar numbers existed elsewhere.

Ben Nimmo, principal investigator at OpenAI, didn't sugarcoat it: "This is what Chinese modern transnational repression looks like. It's not just digital. It's not just about trolling. It's about trying to hit critics of the CCP with everything, everywhere, all at once."

The operation generated millions of posts on Chinese networks and tens of thousands on foreign platforms. All coordinated. All designed to silence dissent abroad.

"It's industrialized," Nimmo said.

The Prime Minister Plot

The same ChatGPT account tried something even more ambitious: influencing Japanese politics.

The user asked ChatGPT to draft a multi-part plan to undermine Japan's incoming prime minister, Sanae Takaichi. The strategy: fan online outrage about US tariffs on Japanese goods and redirect that anger toward Takaichi.

ChatGPT refused to help. But the plan went forward through other channels. In late October 2025, as Takaichi took power, hashtags attacking her and complaining about US tariffs emerged on a popular Japanese graphic arts forum. The pattern matched what the ChatGPT user had described.

One AI service refused to participate. The operation continued anyway. That's the limitation of content moderation: you can kick someone off your platform, but you can't stop them from finding another way.

Not New, Just Visible

None of this is new. Transnational repression (governments targeting their citizens abroad) has been documented for years. What's new is seeing the operational details written down by someone who thought ChatGPT was a secure notepad.

The FBI maintains a dedicated page on transnational repression. Freedom House tracks cases globally. Human rights organizations have documented Chinese "police stations" operating in dozens of countries, pressuring dissidents to return home.

What OpenAI's report adds: tactical specifics. How the calls work. What the forged documents say. How operators coordinate fake obituaries across platforms. The machinery behind the intimidation.

What This Means for Dissidents

If you're a Chinese dissident abroad (or a family member of one, or a journalist covering Chinese politics) you're a potential target. The operation doesn't distinguish between high-profile activists and ordinary people who said the wrong thing online.

Don't trust unsolicited calls from "government officials." Real immigration authorities don't cold-call people to threaten deportation.
Verify any "legal documents" independently. Call the court directly using a number you look up yourself, not one provided in the document.
Report suspicious contact to the FBI. Their tip line: 1-800-CALL-FBI. Their online portal: tips.fbi.gov.
Document everything. Screenshots, call recordings (where legal), email headers. This evidence matters.

The AI Company as Intelligence Service

There's an uncomfortable irony here: OpenAI, a company that sells AI services to anyone who pays, discovered a foreign intelligence operation by analyzing how a customer used ChatGPT.

OpenAI has access to everything you type into ChatGPT. Every prompt. Every conversation. The company says it uses this data responsibly, but "responsible" means different things to different people.

In this case, OpenAI acted as a de facto intelligence service, detecting foreign influence operations the same way NSA or CIA might. That's useful when the target is Chinese transnational repression. It's more complicated when you consider what else they could detect, and who might ask them to.

The same surveillance capability that caught this operative could be used to monitor journalists, activists, whistleblowers, or anyone else who types sensitive information into a chatbot. One person's safety feature is another person's threat.

The Bottom Line

A Chinese law enforcement official kept work notes in ChatGPT. OpenAI's threat team found them. Now we know how transnational repression operates at the ground level: impersonated officials, forged documents, faked deaths, coordinated harassment.

Hundreds of operators. Thousands of fake accounts. Millions of posts. All targeting people whose crime was leaving China and speaking freely.

The user who kept the diary got banned. The operation continues. The tactics they documented are still being used against dissidents right now, by colleagues who were smart enough not to use American AI as a notepad.

That's the thing about exposing one careless operative: it reveals the method, but it doesn't stop the machine.

References

Published: February 27, 2026