Pentagon Staff Still Using DeepSeek – Bloomberg Reports

Bloomberg reveals that Pentagon employees continue to use DeepSeek despite security concerns. 

According to Bloomberg, Pentagon staff have been downloading an early version of the DeepSeek Chinese generative artificial intelligence model onto their workstations since fall 2024. The Pentagon’s IT experts only began partially restricting the app’s usage after it recently gained widespread popularity.

Earlier this month, a Chinese startup introduced its newest open-source AI model, R1. According to the creators, this model has surpassed top products from US developers like OpenAI’s flagship o1 in certain scenarios. Its availability—which enables anyone to download and operate it on their own servers at no cost—has excited the open-source community and led to a sell-off of US tech stocks on Monday.

According to a report by Bloomberg on Thursday, the Defense Information Systems Agency (DISA), responsible for overseeing the Pentagon’s IT networks, took action late Tuesday to block access to the DeepSeek website. This decision came after numerous defense employees had already been using the latest AI chatbot for at least two days. The information was provided by unnamed US defense officials.

Experts at DISA are currently evaluating how extensively employees are directly using DeepSeek through web browsers. When accessed via the company’s website or its popular apps on the Apple App Store and Google Play, it was revealed in DeepSeek’s privacy policy that user data is stored on servers located in China and managed under Chinese law.

Despite the restrictions, thousands of Pentagon employees still use Chinese-trained AI via the authorized web platform Ask Sage. This platform offers access to multiple models, including DeepSeek, which are hosted on its own US-based servers.

According to Bloomberg sources, certain US military personnel installed an earlier version of DeepSeek on their workstations as early as the fall of 2024. At that time, Defense Department security teams reportedly did not recognize any connection to China, and therefore the downloads did not immediately raise concerns.

The increasing attention towards DeepSeek has led the US military to identify and delete code from China-based chatbots installed on employees’ devices. Citing security and ethical concerns tied to the model’s origins, the Navy has prohibited any use of DeepSeek. Meanwhile, although the Air Force hasn’t provided specific guidance regarding DeepSeek, it does restrict using sensitive public information in commercial generative AI systems without appropriate approvals.

Sam Altman, the CEO and co-founder of OpenAI, has recognized DeepSeek R1 as an “impressive model,” especially considering what it offers for its price. The Chinese startup asserts that it created a competitive AI model costing only $6 million without using advanced US chips.

U.S. President Donald Trump, emphasizing the importance of AI development as a key focus for his administration, described DeepSeek’s launch as a “wake-up call” for the American AI sector. Last week, he unveiled Stargate—a plan to construct data centers nationwide intended to support future AI endeavors—aiming at an investment goal of up to $500 billion.