Bloomberg: Pentagon Personnel Using Chinese AI DeepSeek

Reports indicate that US military personnel have been using an early version of a Chinese AI model for several months.

Bloomberg reports that Pentagon staff downloaded an early version of the Chinese startup’s generative AI model, DeepSeek, to their computers beginning in the fall of 2024. The Pentagon’s IT department only attempted to partially restrict DeepSeek’s use after its recent surge in popularity.

Earlier this month, DeepSeek released its open-source AI model, R1, which its developers claim surpasses leading US AI products, including OpenAI’s o1 in some areas. The model’s free and readily available nature has captivated the open-source community and led to a significant drop in US tech stock prices on Monday.

Bloomberg, citing unnamed US defense officials, reported that the Defense Information Systems Agency (DISA), responsible for the Pentagon’s IT infrastructure, blocked access to DeepSeek’s website late Tuesday, after many defense employees had already used the AI chatbot for at least two days.

DISA is currently evaluating the extent of employee use of DeepSeek through web browsers. When accessed via the company’s website or apps on Apple’s App Store and Google Play, DeepSeek’s privacy policy states that user data is stored on servers in China and subject to Chinese law.

Despite these restrictions, thousands of Pentagon employees continue using DeepSeek through Ask Sage, an authorized platform hosting several AI models, including DeepSeek’s, on US-based servers.

Bloomberg sources indicate that some US military personnel downloaded an earlier version of DeepSeek to their workstations as early as fall 2024. At the time, the connection to China was reportedly not immediately apparent to Department of Defense security teams, and the downloads did not initially raise concerns.

The increased interest in DeepSeek has prompted the US military to remove Chinese-origin chatbot code from employee computers. The Navy has banned all use of DeepSeek due to security and ethical concerns. While the Air Force hasn’t issued specific DeepSeek guidance, it prohibits using sensitive information with commercial generative AI systems without proper authorization.

OpenAI CEO and co-founder Sam Altman described DeepSeek R1 as an “impressive model, particularly in terms of what they’re able to deliver for the price.” The Chinese startup claims to have developed this competitive AI model for just $6 million, without access to advanced US chips.

US President Donald Trump, who has prioritized AI development, called DeepSeek’s launch a “wake-up call” for the American AI industry. Last week, he announced Stargate, an initiative to build US data centers for future AI projects, with a projected investment of up to $500 billion.