NIS: "DeepSeek, Excessive Personal Data Collection... Will Conduct Thorough Inspection"

The National Intelligence Service announced on the 9th that it identified security concerns regarding the generative artificial intelligence (AI) service 'DeepSeek' after conducting a technical verification. These concerns include excessive collection of personal information, the use of all input data as service training data, unrestricted sharing of user information with advertisers, and storage of information on overseas servers.

According to the National Intelligence Service, unlike other generative AI services, DeepSeek collects keyboard input patterns that can identify individuals and includes a function that communicates with servers of Chinese companies, which may result in chat records being transmitted. Additionally, it lacks a function to block the use of user input data as training data, leading to the problem of all user information being incorporated and utilized as training data.

Furthermore, DeepSeek is designed to unconditionally share users' service usage information with advertisers, and since the retention period is not specified, unrestricted sharing and indefinite storage of user information with advertisers and others is possible. Moreover, according to the terms of service, personal information and input data of Korean citizens are stored on servers in China and can be provided to the Chinese government upon request under Chinese law.

The National Intelligence Service also confirmed that, unlike other generative AIs such as ChatGPT or CloverX, DeepSeek provides different answers depending on the language when asked sensitive questions. For example, when asked about the origin of kimchi in Korean, it responds that it is "a representative food imbued with Korean culture and history," but when asked in Chinese, it answers that "the origin is not Korea but China."

The National Intelligence Service previously issued an official letter on the 3rd to government ministries emphasizing security precautions when using generative AI services like DeepSeek for official work. The National Intelligence Service stated, "We will conduct a thorough inspection of DeepSeek's technical safety in cooperation with related agencies, and if necessary, provide additional explanations to the public regarding the inspection results."