OpenAI has announced the removal of multiple ChatGPT accounts tied to Chinese and North Korean state-affiliated hackers who were using the platform to refine malware, craft phishing campaigns, and build surveillance tools.
The Findings
In its October 2025 threat report, OpenAI detailed how Chinese-speaking threat actors exploited ChatGPT’s capabilities to accelerate existing cyber operations.
Among them was a cluster called “Cyber Operation Phish and Scripts,” consistent with the activity of groups tracked as UNKDROPPITCH and UTA0388, believed to serve China’s intelligence community.
These actors used ChatGPT for:
- Malware Development: Generating and debugging code for malware families like GOVERSHELL and HealthKick, while exploring automation through other AI systems such as DeepSeek.
- Phishing Campaigns: Producing multilingual phishing emails (Chinese, English, Japanese) targeting Taiwan’s semiconductor industry, U.S. academic institutions, and critics of the Chinese government.
State-Level AI Misuse
OpenAI also banned accounts linked to Chinese government agencies that attempted to use AI for surveillance-related projects, including:
- A “Uyghur Inflow Warning Model” analyzing travel and police data.
- A “Social Media Probe” designed to monitor political or religious content on platforms like X, Facebook, and Reddit.
- Research into identifying critics of the Chinese government and tracking their funding sources.
Mitigation Measures
OpenAI has:
- Disabled all accounts connected to these malicious operations.
- Shared threat indicators with global cybersecurity partners.
- Enhanced internal detection to identify future AI misuse attempts.
Despite the abuse attempts, OpenAI noted that ChatGPT’s built-in safeguards prevented direct generation of malicious code or exploits, limiting attackers to benign snippets.
The Bigger Picture
This is part of a broader effort to protect AI systems from exploitation by authoritarian regimes and cybercriminals. Since February 2024, OpenAI has disrupted over 40 malicious networks.
The company emphasized that while AI is increasingly part of threat actors’ toolkits, its use remains augmentative—not revolutionary—in cyber operations.
