SCIENCE & TECH: DeepSeek app stores user data in China — sparking US security concerns: experts

DeepSeek, the Chinese app that sparked a $1 trillion US market meltdown this week, is storing its fast-growing troves of US user data in China – posing many of the same national security risks that led Congress to crack down on TikTok.

The artificial intelligence chatbot topped the charts in Apple’s App Store and Google’s Play Store on Tuesday. DeepSeek has been downloaded more than 2 million times since its debut on Jan. 15, with most coming in the last three days, according to AppMagic.

While rival chatbots including ChatGPT collect vast quantities of user data, the use of China-based servers by DeepSeek — created by math geek hedge-fund investor Liang Wenfeng — are a key difference and a glaring privacy risk for Americans, experts told The Post.

“What sets this context apart is that DeepSeek is a Chinese company based in China,” said Angela Zhang, a law professor at the University of Southern California focused on Chinese tech regulations.

“This raises the question of whether the collection of data such as IP addresses and keystroke patterns could pose a national security threat,” Zhang added.

DeepSeek’s terms of service disclose that user data is stored “in secure servers located in the People’s Republic of China.” The company also says it automatically collects data on personal information such as “device model, operating system, keystroke patterns or rhythms, IP address, and system language.”

All China-based companies are subject to the Chinese Communist Party’s cybersecurity laws, which mandate that it share data with the government upon request.  

The security risks posed by DeepSeek’s ties to Beijing pushed the U.S. Navy to order members to avoid using the chatbot, CNBC reported Tuesday.

A spokesperson for the Navy confirmed the military branch sent an email to “shipmates” and said it was in reference to the Department of the Navy’s Chief Information Officer’s generative AI policy, according to the outlet.

The Chinese chatbot has also displayed signs of censorship and bias – including refusing to answer prompts about China’s leader Xi Jinping, the Tiananmen Square massacre of 1989, whether Taiwan is a country and if China has committed human rights abuses against Uighurs in Xinjiang.

In some cases, DeepSeek begins generating a response — only to stop itself mid-sentence and write that such prompts are “beyond [its] current scope.”

Data storage in China was a key concern that spurred US lawmakers to pursue a ban of TikTok, which took effect this month after Chinese parent ByteDance failed to divest its stake before a Jan. 19 deadline. President Trump issued an executive order delaying enforcement of the ban.

This week, the House’s bipartisan select committee on China, which led the charge on TikTok, retweeted an image depicting DeepSeek as a “trojan horse.”

Committee chairman Rep. John Moolenaar (R-Mich.) said Congress should “work to swiftly place stronger export controls on technologies critical to DeepSeek’s AI infrastructure.”

“The US cannot allow CCP models such as DeepSeek to risk our national security and leverage our technology to advance their AI ambitions,” Moolenaar said in a statement.

Josh Kushner, whose venture firm Thrive Capital is a major investor in OpenAI, ripped colleagues who were publicly touting DeepSeek, alleging it was built using US technology.

“’Pro America’ technologists openly supporting a Chinese model that was trained off of leading US frontier models, with chips that likely violate export controls, and – according to their own terms of service – take US customer data back to China,” Kushner wrote Monday on X.

A day earlier, Elon Musk tweeted that DeepSeek “obviously” had access to a significant amount of advanced Nvidia chips. ScaleAI’s Alexandr Wang told CNBC that the firm has 50,000 advanced chips it can’t publicly acknowledge due to export controls.

Wedbush analyst Dan Ives said claims that DeepSeek’s model was “built for $6 million with no Nvidia next generation hardware is likely a fictional story.”

DeepSeek did not immediately return The Post’s request for comment. The AI firm does not appear to have a dedicated public relations team yet.

“DeepSeek’s privacy policy, which can be found in English, makes it clear: user data, including conversations and generated responses, is stored on servers in China,” added Adrianus Warmenhoven, a cybersecurity expert at NordVPN.

“Users should consider whether their interactions or uploaded data might inadvertently contribute to machine learning processes, potentially leading to data misuse or the development of tools that could be exploited maliciously,” Warmenhoven added.

Elsewhere, OpenAI’s Sam Altman admitted that DeepSeek was an “impressive model,” but vowed that his firm would “obviously” surpass it with upcoming releases.

By comparison, OpenAI’s GPT-4 cost more than $100 million to train, while Anthropic’s Dario Amodei suggested last year that models currently in development could cost $1 billion or more.

The revelation caused US markets to tank, erasing more than $1 trillion in market value as investors grappled with the possibility that advanced hardware built by Nvidia to power AI models may be less necessary than previously thought.

It also caused many to question whether the US had lost its perceived AI advantage to China.

Billionaire tech investor Marc Andreessen said DeepSeek’s rise was “AI’s Sputnik moment” – a reference to the Soviet Union’s launch of an Earth-orbiting satellite in 1957 that stunned the US and sparked the space race between the two superpowers.