Pentagon Staff Use Chinese AI Model Despite Restrictions

U.S. military personnel have been utilizing an early version of a Chinese startup's AI model for several months now. According to reports, Pentagon staff began downloading this generative artificial intelligence model onto their computer systems in the fall of 2024. This situation arose as the Pentagon's IT experts only decided to limit the use of DeepSeek after seeing a significant increase in its adoption.

This month, DeepSeek launched an open-source AI model named R1, which its developers assert has outperformed several leading AI products from U.S. companies, including OpenAI's flagship model. The model’s availability allows anyone to download it for free, which has excited the open-source community and caused a noticeable drop in U.S. tech stock prices.

The Defense Information Systems Agency (DISA), responsible for managing the Pentagon's IT networks, attempted to block access to DeepSeek's website after many defense employees had already been using the AI chatbot for a couple of days, as reported by unnamed U.S. defense officials. DISA is currently investigating how much DeepSeek is being accessed through web browsers by employees.

DeepSeek’s privacy policies indicate that user data is stored on servers located in China, and any data collected is governed under Chinese law. Despite the initiated restrictions, thousands of Pentagon employees are still accessing DeepSeek via Ask Sage, a sanctioned platform that allows usage of various AI models, including DeepSeek's, on U.S.-based servers.

Some military personnel reportedly began downloading an earlier version of DeepSeek as soon as the fall of 2024, a time at which the connection to China was allegedly not clear to the Defense Department's security teams. This lack of awareness meant that the downloads didn't immediately raise any red flags.

The escalating interest in DeepSeek has prompted the military to find and remove any China-sourced code from devices used by staff. The Navy has outright banned the use of DeepSeek due to concerns regarding security and the ethical implications of its origins. Although the Air Force has yet to provide specific directives regarding DeepSeek, it generally forbids the use of sensitive public information in commercial generative AI systems without the necessary approvals.

OpenAI's CEO and co-founder, Sam Altman, acknowledged the R1 model as highly impressive, particularly for what it delivers at a lower cost. The Chinese startup boasts that it developed a competitive AI product spending just $6 million, without relying on advanced U.S. semiconductor technology.

Former U.S. President Donald Trump has highlighted AI development as a critical focus for his administration and remarked that DeepSeek’s launch serves as a "wake-up call" for the American AI sector. Recently, he announced the Stargate initiative, which aims to establish data centers across the United States to bolster future AI projects, targeting an investment of up to $500 billion.

Pentagon, AI, DeepSeek