OpenAI reveals how Chinese law enforcement exploited ChatGPT to harass dissidents globally

Chinese Law Enforcement Exploits ChatGPT for Cyber Special Operations

OpenAI has exposed a disturbing pattern of malicious uses of artificial intelligence by Chinese law enforcement, detailing how ChatGPT was exploited to coordinate what operatives called "cyber special operations" against political dissidents and foreign leaders 1

1

. Researchers at OpenAI identified an account believed to be based in mainland China that used the AI tool to edit reports and refine plans for large-scale campaigns designed to harass political dissidents living in the US, Japan, and Italy 1

1

. The individual associated with Chinese law enforcement continuously polished updates about their covert influence operations, revealing what Ben Nimmo, principal investigator on OpenAI's intelligence and investigations team, described as a "large-scale, resource-intensive and sustained" effort to suppress dissent 3

3

.

Campaign to Discredit Japan's PM Sanae Takaichi

In mid-October, the user attempted to leverage ChatGPT to design a campaign aimed at undermining Sanae Takaichi, Japan's first female prime minister, after she publicly criticized human rights conditions in Inner Mongolia 2

2

. The plan to discredit Japan's PM involved six elements: posting and amplifying negative comments about Takaichi on social media platforms, sending complaints to Japanese politicians using fake email accounts posing as foreign residents, and accusing her of far-right leanings 3

3

. However, ChatGPT refused to help the individual refine the campaign 2

2

. The user later returned weeks afterward to update a report indicating the campaign proceeded anyway, likely using locally hosted Chinese AI models 3

3

.

Transnational Repression Targets Dissidents Worldwide

The scope of transnational repression operations was massive, targeting 300 social media platforms outside of China with tens of thousands of posts from thousands of accounts 1

1

. Chinese operatives fed ChatGPT detailed information about forging fake documents, including materials seemingly from a US county court, which they planned to use to manipulate social media platforms into removing anti-CCP content 1

1

. Impersonation emerged as a common tactic, with Chinese officials posing as US immigration officials to intimidate US-based dissidents by claiming their statements violated the law 1

1

.

Industrialized Harassment of Individual Critics

Five separate fake Bluesky accounts were created for Hui Bo, a prominent Chinese dissident living in California, while similar fake accounts targeted Teacher Li, a key figure in the 2022 anti-lockdown protests now residing in Italy 1

1

. In one particularly cruel campaign, operatives created a fake obituary and fabricated pictures of gravestones claiming political dissident Jie Lijian had died, then spread this content widely across social media 1

1

. AI was also deployed to generate fake evidence supporting takedown claims directed against pro-Taiwan X accounts, backing allegations of activity that violated platform rules 1

1

. "It's not just digital, it's not just about trolling, it's industrialized," Nimmo told reporters. "It's about trying to hit critics of the (Chinese Communist Party) with everything, everywhere, all at once" 3

3

.

AI Misuse Extends Beyond State-Backed Groups

OpenAI's latest report on disrupting AI misuse also identified other malicious actors exploiting ChatGPT 2

2

. The company found romance scams targeting Indonesians, where Cambodia-based scammers used ChatGPT to create marketing materials for fake online dating services 3

3

. Content farms linked to Russia employed the chatbot to translate social media comments in Spanish for operations targeting Argentina 3

3

. These campaigns had measurable impact, with many targets losing followers, reducing their activity, or deleting their accounts entirely, though OpenAI noted that only a relatively slim proportion of the network's posts achieved significant attention 1

1

. The disclosure comes as questions about AI guardrails intensify, with rival Anthropic recently loosening its commitment to such safeguards 2

2

.