South Korea’s Ministry of the Interior and Safety and the National Intelligence Service have issued warnings to government agencies about using generative AI tools like DeepSeek. (Image courtesy of Yonhap)
SEOUL, Feb. 10 (Korea Bizwire) — Privacy watchdogs and government agencies are raising alarms about DeepSeek, a Chinese artificial intelligence platform, amid reports of extensive data collection practices. However, tech enthusiasts and small businesses are increasingly turning to the platform’s open-source models as an alternative.
According to DeepSeek’s privacy policy released on February 9, the platform collects a broad range of user data, including basic personal information like names and birthdates, as well as more granular data such as IP addresses, unique device identifiers, and keystroke patterns. This information is transmitted in real-time to the company’s servers in China when users access DeepSeek through its web interface or mobile application.
The company also disclosed that it receives additional user activity data from advertisers and partners about users’ behaviors on external websites, apps, and app marketplaces. Security experts note that this level of data collection exceeds that of industry leaders like OpenAI’s ChatGPT and Google’s Gemini.
Unlike ChatGPT, which allows users to opt out of data collection for model training and conversation memory, DeepSeek offers no such choices to its users. Ivan Tsarynny, CEO of Feroot Security, a Canadian cybersecurity firm, claimed in an ABC interview that his team discovered hidden code in DeepSeek that transmits user data to China Mobile, a state-controlled telecommunications company.
In response to these concerns, South Korea’s Ministry of the Interior and Safety and the National Intelligence Service have issued warnings to government agencies about using generative AI tools like DeepSeek. Major Korean companies, including Kakao, have also begun blocking access to the platform.
However, tech-savvy users have found a workaround by downloading and running DeepSeek’s open-source models locally. While the company’s main model, DeepSeek V3, requires substantial computing power with 671 billion parameters and a 404GB file size, its smaller R1 model ranges from 1.5 billion to 70 billion parameters and requires only 4.7GB of storage.
Notably, while DeepSeek’s online version blocks or refuses to respond to queries about sensitive topics like Chinese President Xi Jinping and the Tiananmen Square protests, the downloadable models reportedly operate without such restrictions.
The trend has caught the attention of major tech companies. Nvidia, Microsoft, Intel, AMD, and Amazon Web Services have begun incorporating DeepSeek’s open-source models into their offerings. In South Korea, AI company Wrtn launched a free “secure service” on February 4 that runs DeepSeek models on separate cloud infrastructure through KakaoTalk channels.
Medical AI company Deepnoid announced on February 5 that it had integrated the DeepSeek R1 model into its DEEP:GEN platform to improve inference and generation accuracy. The company emphasized that user data remains secure as the system operates in an isolated cloud environment, physically separated from DeepSeek’s infrastructure.
“While there are valid concerns about the app version’s security, DeepSeek’s impact on on-device AI development could be significant,” said an IT industry insider who reported successful implementation of the model on a personal workstation.
Kevin Lee (kevinlee@koreabizwire.com)
