DeepSeek AI chatbot under scrutiny over data security and language-based bias
South Korean intelligence reported that DeepSeek's AI chatbot provides different answers to sensitive questions depending on the language used, often favoring China's claims on disputed cultural issues. Multiple governments and organizations have banned or sought to ban the platform due to security concerns.
What happened
South Korea's National Intelligence Service (NIS) reported that DeepSeek's AI chatbot provides different responses to sensitive questions depending on the input language. When asked about kimchi in Korean, the chatbot states it is a distinctive Korean food with Korean cultural roots; in Chinese, it claims kimchi originated in China; in English, it says kimchi is related to Korea.
Similarly, when asked about the Dano festival in Korean, DeepSeek identifies it as a Korean traditional festival, but in Chinese and English it identifies it as a Chinese traditional festival. According to the NIS, kimchi and the Dano festival are widely recognized as Korean.
U.S. senators Darin LaHood and Josh Gottheimer have prepared legislation to ban federal employees from using DeepSeek on government devices due to national security concerns. If passed, government agencies would have 60 days to develop standards and guidance to remove DeepSeek and other applications from High Flyer on government devices.
Reported impact
Multiple South Korean organizations and businesses have banned DeepSeek due to data security concerns. Australia, Italy, and some organizations have issued similar bans.
DeepSeek's advanced R1 model, released by the more than one-year-old Chinese startup, has prompted U.S. concerns about American technological advancement in AI.
Index analysis
The chatbot's language-dependent responses on disputed cultural issues suggest potential bias in the model's training or design, with outputs favoring Chinese claims when prompted in Chinese.
The reported data security concerns and government bans indicate broader worries about potential data access and surveillance risks associated with Chinese-based AI services, similar to concerns previously raised about ByteDance's TikTok.
Governance controls
The U.S. proposed legislation would require federal agencies to establish removal standards within 60 days, applying restrictions similar to those implemented for TikTok.
Classification
Sources and evidence
Sources not yet published
This record was researched and written by the Index from original reporting. Its sources are still being verified by an editor, and they are listed here once that check is complete. Until then the record carries no reviewer.