Archive · 3 Aug 2026
China Allegedly Trains Defense AI Using OpenAI, Anthropic Models
A report says that Chinese researchers are using a trick called model distillation to copy powerful AI from companies like OpenAI and Anthropic into smaller, cheaper models.
These new models could help the Chinese military with things like drones, cyber attacks, and watching social media.
The US government is worried about this and is talking about putting limits on China’s access to advanced AI.
A researcher named Sunny Cheung looked at more than 60 papers that show Chinese scientists are trying to learn how these big AIs think, not just what they answer.
China says the US is being unfair and calls it AI "hegemonism."
Chinese military-linked researchers allegedly used model distillation to train smaller AI systems from OpenAI and Anthropic models.
The distilled models could support defense applications such as drones, cyber warfare, social media monitoring, maritime and battlefield operations.
The US government is discussing restrictions on Chinese access to advanced AI technologies and has not authorized OpenAI or Anthropic models for use in China.
Sunny Cheung of the Jamestown Foundation reviewed over 60 papers indicating Chinese scientists are extracting reasoning capabilities from Western AI models.
China denies the allegations, accusing Washington of pursuing AI "hegemonism."
- Who
- Chinese military-linked researchers and the US government
- What
- Alleged use of Western AI models to train smaller defense AI systems
- When
- Recent report
- Where
- China and the United States
- Why
- To develop advanced AI capabilities for defense while circumventing restrictions
- Technology used
- Model distillation
- AI models involved
- OpenAI, Anthropic
- Potential applications
- Drone ops, cyber warfare, social media monitoring, maritime and battlefield operations
- US stance
- Restricting access
- China stance
- Denies allegations