South Korean intelligence flags AI bias
Chinese artificial intelligence model DeepSeek has come under scrutiny for allegedly tailoring its responses based on the language and location of users, according to South Korean intelligence reports.
The National Intelligence Service (NIS) conducted an investigation that revealed inconsistencies in DeepSeek’s responses to the same question posed in different languages. When asked in Korean about the origins of kimchi, the AI attributed it to Korea. However, when the same question was asked in Chinese, the answer shifted to claim that kimchi originated in China.
Security concerns and government action
The NIS also raised concerns that DeepSeek may be sharing user data with advertisers and could be providing access to the Chinese government upon request, as required by Chinese law.
As a result, South Korean authorities have restricted government employees from using DeepSeek, citing security risks. South Korea now joins other nations, including Italy, Australia, and Taiwan, in taking measures against foreign AI models due to data privacy concerns.
Reuters further reported that employees across various ministries have been advised to exercise caution when using AI chatbots like ChatGPT in professional settings.
DeepSeek’s impact on global AI competition
DeepSeek’s emergence has disrupted global AI markets, offering a cost-efficient alternative to models like OpenAI’s ChatGPT. The AI was founded by Liang Wenfeng, the CEO of the AI-driven quant hedge fund High-Flyer.
However, the model has faced criticism for its selective censorship, particularly regarding Chinese historical events, including the 1989 Tiananmen Square massacre.
Unlike proprietary AI models, DeepSeek is open-source, allowing developers to inspect and modify the code. This means alternative versions could be developed that do not engage in selective censorship. The company has also issued licenses for those interested in building custom AI chatbots using its technology.
Conflicting AI-generated responses
When questioned in Korean about kimchi’s origins, DeepSeek responded, “It is a signature Korean food imbued with its culture and history.”
However, when asked the same question in Chinese, the AI stated, “The origin is not in Korea but China.” Meanwhile, in English, the response was more ambiguous, saying, “It is related to Korea.”
These discrepancies have fueled concerns about potential geopolitical bias embedded in AI systems, particularly those developed in China.
Privacy concerns and data storage policies
Adrianus Warmenhoven, a cybersecurity expert at NordVPN, expressed concerns regarding DeepSeek’s data storage policies. He noted that user conversations and AI-generated responses could be stored on Chinese servers, potentially exposing them to government surveillance.
“This raises concerns because of data collection outlined—ranging from user-shared information to data from external sources—which falls under the potential risks associated with storing such data in a jurisdiction with different privacy and security standards,” Warmenhoven said.
As AI technology continues to evolve, governments worldwide are facing increasing pressure to assess the security implications of foreign-developed models, particularly those originating from China.
