China's military seeks DeepSeek and Qwen models for cyber training
House committees revealed that the Chinese military is procuring specific AI models for command systems and cyber ranges, sparking calls for broader U.S. export controls.
The short version
- The PLA sought DeepSeek and Qwen models for an intelligent command information system and a cyber range training environment.
- Military procurement notices required these models to run on isolated internal networks without internet access.
- Personnel from a PLA cyber intelligence unit visited the headquarters of Zhipu, a Beijing-based AI company, in August.
- U.S. lawmakers are now pushing to expand federal restrictions beyond DeepSeek to include other Chinese model families.
Why it matters
This shift indicates that the U.S. government may soon treat all Chinese AI models as significant security risks. Companies currently using these models for internal tools or customer support may face new regulatory hurdles or deployment restrictions.
Military procurement of AI
House committees recently uncovered records showing that China's military is actively seeking to integrate specific artificial intelligence models into its infrastructure. The People's Liberation Army (PLA) has expressed interest in using DeepSeek and Alibaba's Qwen models for various internal systems. These findings come from a comparison of military purchasing records and documents from American companies.
The military specifically requested servers capable of running these models to power an intelligent command information system. Additionally, a procurement notice from July 2 specified the use of DeepSeek models for a cyber range training environment. These environments are used to simulate and practice digital warfare and cybersecurity maneuvers.
The scope of these requests appears to be broad and varied across different military functions. For example, a notice from October 2024 sought Zhipu's GLM models to run on a military office network. Another notice from July 2 called for DeepSeek or Qwen to assist in generating and checking medical records for the military.
Security and local deployment
A key technical requirement in these military procurement notices is the need for local execution. The models are intended to run on isolated internal military networks that do not have any connection to the public internet. This setup ensures that the data processed by the AI remains within the military's private infrastructure.
The military also specified that these models should be capable of running on standard military office networks. By requiring local hosting, the military can utilize the capabilities of large language models without the risks associated with cloud-based processing. This allows for the use of advanced AI tools while maintaining a closed loop for sensitive information.
The investigation also highlighted physical interactions between the military and AI developers. People associated with the 1st Bureau of the PLA Cyberspace Force visited the Beijing headquarters of Zhipu in August. This visit suggests a direct line of communication between military intelligence units and the companies developing the underlying technology.
Impact on U.S. companies
The investigation into Chinese AI models has already touched several prominent American companies. Airbnb informed investigators that it has been running customized versions of the Qwen models on infrastructure located within the United States. These models were specifically deployed to handle customer support tasks for the lodging marketplace.
DoorDash was also included in the House inquiry in July regarding its use of Chinese AI models. The committee's investigation compared how these models are used by private American firms against how they are being procured by the Chinese military. This comparison helps lawmakers understand the dual-use nature of these specific AI model families.
The findings raise significant questions about how these models are secured when deployed by private entities. Because the same models used by American companies for commercial purposes are being sought by the PLA for military use, the security implications are heightened. This creates a complex landscape for U.S. businesses that rely on these specific technologies for their daily operations.
Shifting federal restrictions
Current federal restrictions in the United States have primarily focused on DeepSeek. However, the new findings from the House committees are prompting a shift in policy. Lawmakers are now urging Washington to broaden these restrictions to include a wider family of Chinese AI models, such as Qwen and GLM.
This move marks a transition toward treating the entire ecosystem of Chinese AI models as a collective security concern. Instead of targeting a single company or model, the government may move toward a broader prohibition. This would affect any company that currently integrates Chinese-made AI into their internal tools or customer-facing services.
The policy landscape is expected to become more restrictive for organizations deploying open AI models. As the government moves to cover more model families, businesses will need to be more diligent in their procurement. Any new federal rules could fundamentally change which models are legally allowed for deployment in the United States.
Next steps for organizations
The evolving situation suggests that the regulatory environment for AI is becoming more complex and less predictable. Organizations that currently use open AI models should begin documenting exactly where each model runs. It is important to identify whether a model is hosted locally, on a private cloud, or on a public internet-connected server.
Companies should also maintain clear records of who has access to these models and what types of data they process. Since the government may soon expand restrictions to cover more than just DeepSeek, having this documentation ready will be essential. This proactive approach can help businesses pivot quickly if certain models are suddenly restricted by new federal policies.
The ongoing inquiry by House committees suggests that more information regarding Chinese AI procurement will emerge soon. As the government continues to investigate the links between Chinese AI developers and military interests, the scope of restrictions will likely expand. Businesses should stay informed on these policy shifts to ensure their AI strategies remain compliant with federal law.
What to watch
- Whether the U.S. government officially expands export restrictions to include the Qwen and GLM model families.
- Further details on the specific capabilities of the PLA's 'intelligent command information system' powered by AI.
Try it
Review your company's AI inventory to see if any Chinese-made models, such as Qwen or DeepSeek, are currently being used for customer support or internal data processing.
Sources and links
- ReportingWisevoter: the article we sourced
Read the original at Wisevoter
First reported by Wisevoter. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.