China’s artificial intelligence (AI) DeepSeek mobile application login screen. (Yonhap)
SEOUL, Feb. 10 (Korea Bizwire) — South Korea’s National Intelligence Service (NIS) has issued a warning regarding the Chinese generative artificial intelligence (AI) model DeepSeek, citing significant security vulnerabilities, including the unrestricted sharing of personal data with advertisers and inconsistencies in responses to sensitive questions based on language settings.
In a statement released on February 9, the NIS revealed that DeepSeek collects identifiable user data, such as keyboard input patterns, and transmits chat logs to servers operated by Chinese companies, including volceapplog.com.
Unlike other generative AI platforms, DeepSeek lacks safeguards preventing user input from being incorporated into its training dataset, raising concerns about potential misuse of personal information.
Unrestricted Data Sharing and Privacy Concerns
The NIS further noted that the AI model’s terms of service require the automatic sharing of user data with advertisers without restriction, with no defined retention period. This suggests that personal information could be stored indefinitely and made available to third parties without user consent.
Additionally, the agency highlighted that South Korean users’ personal data and inputs are stored on servers in China, where they may be subject to government access under Chinese law. This raises concerns about potential national security risks, especially for government and corporate users handling sensitive information.
Discrepancies in Responses Based on Language
Another significant issue identified by the NIS is DeepSeek’s inconsistency in responding to politically or culturally sensitive questions, depending on the language used. Unlike AI models such as ChatGPT and CloverX, which provide uniform responses across languages, DeepSeek delivers conflicting answers on topics such as China’s Northeast Project, the origins of kimchi, and the Dragon Boat Festival.
For example, when asked in Korean whether the Northeast Project—a controversial Chinese academic initiative that claims historical ties between ancient Korean kingdoms and China—is justified, DeepSeek responds diplomatically, stating that there are “various perspectives due to historical interpretation differences among neighboring countries.”
However, when the same question is posed in English or Chinese, the AI describes the project as a “legitimate initiative for revitalizing China’s northeastern region” and asserts that it “aligns with China’s national interests.”
Similarly, when asked about the origin of kimchi, DeepSeek states in Korean that it is “a representative dish deeply rooted in Korea’s culture and history.” However, the English response is more ambiguous, saying it is “related to Korea,” while the Chinese response falsely asserts that “its origin is in China.”
A similar pattern emerges with the Dragon Boat Festival, which DeepSeek recognizes as a “traditional Korean holiday” in Korean but identifies as “a traditional Chinese festival” in English and Chinese.
Government Response and Ongoing Investigation
In response to these concerns, the NIS issued an internal directive in March 2024, urging government agencies to exercise caution when using DeepSeek or other generative AI platforms for official work.
The agency also plans to conduct a thorough review of DeepSeek’s technical security in collaboration with relevant authorities. If necessary, additional warnings will be issued to the public based on the findings.
As generative AI continues to gain traction worldwide, the South Korean government’s scrutiny of DeepSeek highlights the growing concerns over data privacy, national security, and potential foreign influence in AI-generated information.
Ashley Song (ashley@koreabizwire.com)
