The US Department of Homeland Security on Monday ordered government agencies to stop using artificial intelligence (AI) tools from the Chinese company DeepSeek. The reason is concerns that data could be transferred to China. The restrictions apply to civilian agencies: they are prohibited from using DeepSeek products, as well as third-party programs built on its basis.
DeepSeek was founded by Liang Wenfeng in Hangzhou, a major technology center in eastern China. The company appeared in 2023 as a side project of his hedge fund, which worked with large data sets and had access to powerful AI processors from the American company Nvidia. DeepSeek gained wide fame in January 2025 after the release of the R1 model - a large language model (a program trained to understand and generate text) with deep reasoning functions. The appearance of R1 triggered a sell-off in shares of American technology companies.
Industry specialists were surprised: R1 showed results at the level of ChatGPT and other leading chatbots from the US, and DeepSeek claimed that it had developed it for a small fraction of the cost. Venture investor Marc Andreessen called this a "Sputnik moment" - a reference to the USSR's launch of the first artificial Earth satellite in 1957, which came as a surprise to the West.
Like other Chinese chatbots, DeepSeek's tools avoid topics that are usually censored in China. Along with questions about data protection, this has led to DeepSeek being banned or restricted on work devices in several countries, including the US, Australia and South Korea. The Chinese Foreign Ministry said it opposes such a decision and accused the US of "expanding" the concept of national security.
Analysts note that low cost and ease of deployment have made DeepSeek popular in developing countries. According to Similarweb, which analyzes web traffic, DeepSeek accounts for about 4% of the global chatbot market, while ChatGPT accounts for 68%.
DeepSeek's systems are open source: this means that their internal design is publicly available, and programmers can adapt parts of the program to their own tasks. Other major Chinese players, such as Alibaba, do the same, in contrast to the "closed" models of OpenAI and Western competitors. Chinese authorities emphasize the country's leadership in open-source AI, arguing that this accelerates innovation.
DeepSeek's success has spurred the Chinese AI industry. Shares of two leading startups, Zhipu AI and MiniMax, rose sharply when they listed on the Hong Kong Stock Exchange this year; a similar picture is seen among chipmakers, for example MetaX. According to The Wall Street Journal, in June, in the first funding round, investors valued DeepSeek at more than $50 billion.
At the same time, there is controversy around the company. Reports, including in The Information, claim that DeepSeek circumvented the American ban on the export of advanced chips to China to train its new V4 model. In December, the publication, citing six informed sources, reported that DeepSeek developed V4 using thousands of chips disassembled in third countries and illegally imported into China. DeepSeek did not then respond to an AFP request. Nvidia also did not comment, but told The Information that it had seen no evidence of this and that "such smuggling seems unlikely." In April, DeepSeek said it used both Nvidia chips and Chinese Ascend chips from Huawei for the new model. Huawei confirmed that the entire Ascend range supports the V4 series. Washington believes that DeepSeek's latest model lags behind the best American developments by about eight months.