Seoul: The Chinese artificial intelligence model DeepSeek has attracted attention for providing inconsistent answers to culturally sensitive questions, such as the origin of kimchi, based on the language used. According to Seoul’s National Intelligence Service (NIS), when asked in Korean, DeepSeek identified kimchi as a Korean dish, but when queried in Chinese, it stated that kimchi is of Chinese origin.
According to Yonhap News Agency, the NIS has raised concerns about DeepSeek’s data practices, highlighting that the AI service grants advertisers unrestricted access to users’ personal data. This data can also be handed over to the Chinese government upon request, as per Chinese law. The NIS’s assessment of DeepSeek comes amid rising security concerns, which have led South Korean authorities to block access to the service.
The NIS further reported that DeepSeek’s responses vary significantly depending on the language. For instance, when asked in Korean about the controversial Northeast Project, DeepSeek noted “various perspectives” among countries. However, in English or Chinese, it termed the project a “legitimate initiative” aligned with China’s national interests. Similarly, regarding South Korea’s Dano festival, DeepSeek described it as a Korean tradition in Korean, but labeled it a Chinese holiday in Chinese and English.
In contrast, other AI services like OpenAI’s ChatGPT and South Korea’s Naver-developed CloverX provided consistent answers across languages, the NIS noted. Additionally, DeepSeek was found to collect users’ keyboard input patterns, potentially allowing for the identification of individuals. The AI stores user inputs indefinitely, shares data with advertisers without restrictions, and lacks clarity on data retention periods, raising privacy concerns.
The NIS highlighted that DeepSeek’s terms of service suggest the possibility of personal data and input data being accessible to the Chinese government. Following these findings, the NIS has issued warnings to all government agencies in South Korea to exercise caution when using generative AI tools like DeepSeek.
The agency has announced plans to conduct an in-depth review of DeepSeek’s technological reliability and intends to provide further public clarifications on its findings if necessary.