China Taps US AI Models for Defence Systems
· news
The Shadow Model: China’s Stealthy Use of US AI Research
The quiet infiltration of American artificial intelligence research into China’s defense systems sends a chilling message to policymakers in Washington. A review of Chinese academic papers and patents by Reuters reveals that military researchers are using outputs from leading US AI models, including those developed by OpenAI and Anthropic, to train their own domestic AI systems for surveillance, cyber warfare, and tactical decision-making.
This trend is not merely intellectual curiosity; it’s a strategic move to bypass export controls imposed by the US on advanced chips and other strategic technologies. Chinese researchers are using “model distillation” to create smaller, specialized models that can be deployed locally without the enormous computing requirements needed to build frontier AI systems from scratch. Model distillation is an industry practice, but its widespread use by Chinese military institutions raises concerns about unauthorized extraction of capabilities from American AI models.
The issue has emerged as a major flashpoint ahead of US-China talks on AI governance and safety. What’s striking is not just the scale or scope, but also the strategic implications. By tapping into the research of their American counterparts, Chinese military scientists are essentially capturing the reasoning steps behind Western models to adapt them for their own purposes. This challenges Washington’s efforts to restrict Beijing’s access to advanced technologies.
The Chinese government has pushed back against accusations of intellectual property theft and US “hegemonism” in AI research. While some American firms have engaged in similar practices, the scale and context are far more significant here. China is leveraging its unique strengths – including vast computing resources and an appetite for technological innovation – by embracing model distillation as a shortcut to developing specialized systems.
This trend should be viewed through the lens of broader global politics. The use of US AI models in Chinese defense systems highlights the challenges of regulating cutting-edge technologies across national borders. As AI advances at breakneck speed, policymakers must grapple with the tension between promoting scientific collaboration and safeguarding intellectual property rights.
Experts caution about model distillation’s limitations – distilled models are less capable and pose security risks – but Chinese researchers are racing to develop defense measures to mitigate these vulnerabilities. The implications of this development will be far-reaching; as AI becomes increasingly integral to military operations, the stakes for policymakers and industry leaders are higher than ever.
The “shadow model” has become a potent tool in Beijing’s arsenal. Coined by Jamestown fellow Sunny Cheung to describe the use of Western models for Chinese purposes, it represents a new era of technological competition between Washington and Beijing. The question now is how Washington will respond to this challenge, and what it means for the future of AI governance and international cooperation.
Reader Views
- ADAnalyst D. Park · policy analyst
The irony is that China's exploitation of US AI research highlights the flaws in our own export control policies. By allowing model distillation, we're inadvertently facilitating the spread of sensitive technologies to adversaries. The focus should shift from accusing Beijing of IP theft to reforming our own practices and strengthening partnerships with trusted allies to counter this trend. What's needed is a more nuanced approach that balances innovation with security, rather than simply trying to restrict Chinese access to cutting-edge research.
- RJReporter J. Avery · staff reporter
The reliance on US AI models by Chinese military researchers is a wake-up call for policymakers, but let's not overstate the novelty of this trend. China has been quietly gathering and adapting cutting-edge research from the West for years. The real concern should be how these distillations are being used in practice - not just in theoretical applications, but in actual operations on the ground. Are we seeing early signs of Chinese AI systems outperforming their American counterparts? If so, this development would require a reevaluation of US export controls and defense strategies.
- CSCorrespondent S. Tan · field correspondent
The shadow model phenomenon is far more insidious than mere intellectual property theft. It's about China's military leveraging the collective efforts of American AI research to bolster its own capabilities. What we're not being told is how these adaptations will impact existing US defense systems, particularly those reliant on proprietary software or hardware. The question is, are we inadvertently feeding our adversaries the keys to our own vulnerabilities?