US military personnel have reportedly utilized an early version of the Chinese startup’s AI for months
Pentagon staff have been downloading an early version the Chinese startup’s generative artificial intelligence model onto their workstations since the fall of 2024, according to Bloomberg. The Pentagon’s IT experts only moved to partially block DeepSeek usage after its recent surge in popularity.
Earlier this month, DeepSeek unveiled an open-source AI model, R1, which its creators claim has outperformed leading products from US developers, including in some cases OpenAI’s flagship o1. The model’s accessibility – allowing anyone to download and run it on their own servers for free – has stirred the open-source community and triggered to a sharp decline in US tech stock prices on Monday.
The Defense Information Systems Agency (DISA), which manages the Pentagon’s IT networks, moved to block access to the Chinese startup’s website late Tuesday after many defense employees had already used the latest AI chatbot for at least two days, Bloomberg reported, citing unnamed US defense officials.
DISA experts are now assessing the extent of employees’ direct use of DeepSeek through web browsers. When accessed via the company’s website or its popular apps on the Apple App Store and Google Play, DeepSeek’s privacy policy indicated that user data is stored on servers in China and governed under Chinese law.
Despite the restrictions, thousands of Pentagon employees continue to use the Chinese AI through Ask Sage, an authorized platform that hosts multiple models, including DeepSeek’s, on its own US-based servers.
Some US military personnel also downloaded an earlier version of DeepSeek onto their workstations as early as the fall of 2024, according to Bloomberg sources. At the time, the connection to China was allegedly “not evident” to Defense Department security teams, and the downloads did not raise immediate concerns.
The growing interest in DeepSeek has prompted the US military to locate and remove code from China-origin chatbots on employees’ machines. The Navy has banned any use of DeepSeek, citing security and ethical concerns related to the model’s origins. The Air Force has not issued specific guidance on DeepSeek but prohibits the use of sensitive public information in commercial generative AI systems without proper approvals.
OpenAI CEO and co-founder Sam Altman acknowledged DeepSeek R1 as an “impressive model, particularly in terms of what they’re able to deliver for the price.” The Chinese startup claims it developed a competitive AI model without access to cutting-edge U.S. chips for just $6 million.
US President Donald Trump, who has designated AI development as a top priority for his administration, called DeepSeek’s launch a “wake-up call” for the American AI industry. Last week, he announced Stargate, an initiative to build data centers across the US to serve as the backbone for future AI projects, with an investment target of up to $500 billion.
