China's Military Seeks DeepSeek Models for Cyber Training
House committees say Chinese military procurement records and US company documents show PLA interest in DeepSeek and other Chinese AI models.
Updated on Nov. 21, 2026 in Artificial Intelligence

Live Poll
Do you trust Chinese-made AI tools with your personal data?
House committees reported that China's military specified DeepSeek models for a cyber range training environment and sought servers able to run DeepSeek and Alibaba's Qwen for an intelligent command information system. The inquiry also found people associated with a PLA cyber intelligence unit visited Zhipu's Beijing headquarters, most recently in August.
Why it matters
The findings extend concerns beyond DeepSeek to a family of Chinese AI models now used by American companies, raising questions about how those models are deployed and secured. Lawmakers have urged Washington to broaden federal restrictions beyond DeepSeek.
Procurement notices required the models to run locally, including on an isolated internal military network without internet access and on a military office network. A July 2 notice called for DeepSeek or Qwen to help generate and check medical records.
The players
DeepSeek
A Chinese artificial intelligence company whose open models have drawn US federal restrictions and attention from lawmakers.
Zhipu
A Beijing-based AI company that develops the GLM family of large language models.
Alibaba
A major Chinese technology conglomerate that develops the Qwen family of AI models.
Airbnb
A US online lodging marketplace that told investigators it ran customized Qwen models on US-based infrastructure.
DoorDash
A US food delivery company whose use of Chinese AI models was added to the House inquiry in July.
The details
The House committees' joint investigation compared military purchasing records with documents from American companies using the same Chinese model families. Airbnb told investigators it ran customized Qwen models on US-based infrastructure for customer support.
Timeline
October 2024: A military procurement notice sought Zhipu GLM models to run on a military office network.
January 2026: The military sought servers able to run DeepSeek and Qwen for an intelligent command information system.
April: The House inquiry began with letters to Airbnb and Anysphere.
July 10: A military procurement notice specified DeepSeek for a cyber range.
August: People associated with the 1st Bureau of the PLA Cyberspace Force made the most recent visit to Zhipu's Beijing headquarters.
Roadmap
Federal restrictions so far have centered on DeepSeek, leaving other Chinese AI model families outside the same limits. The House committees' push to extend those restrictions marks a shift toward treating the wider family of Chinese models as a security concern.
Companies using Chinese AI models for customer support or internal tools may need to review where those models run and what data they touch. Any broadened federal restrictions could change which models businesses are allowed to deploy.
The takeaway
Organizations deploying open AI models should document where each model runs and who can access it, since the policy landscape may shift to cover more than one model family.
Further reading
For more on the policy debate, see Artificial Intelligence coverage across the site.
Live Poll
Do you trust Chinese-made AI tools with your personal data?






