China Used ChatGPT to Target Japan's PM, OpenAI Reveals

A Window Into Chinese Information Warfare

OpenAI's latest threat intelligence report, released this week, has lifted the veil on a previously undisclosed Chinese state-linked operation that attempted to weaponize the ChatGPT chatbot against Japan's first female prime minister. The disclosure offers an unusually detailed glimpse into how authoritarian governments are experimenting — sometimes clumsily — with commercial AI tools to conduct overseas influence operations.

The Target: Japan's New Prime Minister

According to OpenAI's findings, analysts identified a troubling planning session in mid-October 2025, when a ChatGPT account associated with a Chinese law enforcement agency asked the model to help design a covert propaganda campaign against Sanae Takaichi. Takaichi, who has since become Japan's first female prime minister, had publicly criticized the Chinese Communist Party for human rights abuses in Inner Mongolia — a statement that apparently triggered the operation.

The account requested that ChatGPT craft negative content for social media, generate fake complaints from supposed foreign residents, and route those complaints to other Japanese politicians — all aimed at discrediting Takaichi before she could consolidate power.

ChatGPT Refused. The Campaign Went Ahead Anyway.

OpenAI says its safeguards functioned: ChatGPT declined to assist with the propaganda requests. But the story does not end there. Weeks later, the same account returned — this time to upload a report indicating the smear operation had proceeded regardless, almost certainly using domestically developed Chinese AI models that carry fewer content restrictions.

"This operation revealed a lot about China's strategy for covert influence operations and transnational repression," said Ben Nimmo, principal investigator on OpenAI's intelligence and investigations team.

The pivot to homegrown AI after ChatGPT's refusal underscores the limits of any single platform's content policies as a bulwark against state-sponsored disinformation.

Industrial Scale: Hundreds of Operatives, Thousands of Accounts

While only one account interacted with ChatGPT, OpenAI's wider investigation found the underlying Chinese network to be large-scale, resource-intensive, and sustained. The operation reportedly deploys hundreds of human staff managing thousands of inauthentic accounts spread across multiple social media platforms. Tactics include mass posting, generating fake complaints against dissident accounts, forging documents, and impersonating U.S. officials to intimidate overseas Chinese critics.

Part of a Broader Misuse Picture

The full threat report documents a range of ChatGPT misuse beyond state actors. Scammers in Southeast Asia used the platform to run romance fraud schemes targeting Indonesian men through a fake dating service. Others impersonated law firms and U.S. law enforcement agencies to deceive fraud victims. Several accounts attempted to gather intelligence on American infrastructure, while some actors sought guidance on face-swapping software for deception campaigns.

What This Means for AI Governance

The incident crystallizes a fundamental tension in the AI industry: powerful tools that refuse harmful requests on Western platforms can be replicated by state actors using domestic alternatives operating under different rules. As generative AI becomes central to geopolitical competition, analysts say the episode strengthens the case for international frameworks setting baseline standards for AI model behavior.

For now, OpenAI's decision to publish its findings in detail stands as one of the few mechanisms for public accountability in what is otherwise a largely invisible domain — a domain that is rapidly becoming one of the defining arenas of 21st-century statecraft.