Chinese startup DeepSeek recently made waves in the AI community by releasing its open-source model, DeepSeek-R1. This model uses pure reinforcement learning, challenging the notion that only large-scale training with powerful chips can lead to high-performing AI. However, concerns emerged about the transmission of personal user data to China as part of DeepSeek’s privacy policy.
In its privacy policy, DeepSeek notes that it collects information from users in various ways, including account setup information and usage data. The information collected is stored in secure servers located in China and may be shared with law enforcement agencies and public authorities. This has raised alarms about security and privacy implications, including profiling individuals, leakage of sensitive business data, and cyber surveillance campaigns.
The privacy policy specifically applies to DeepSeek’s own services using the R1 model in the cloud. If users sign up for these services, there is a high likelihood that their data has been sent to and stored in China. The open-source DeepSeek-R1 model itself, however, does not involve data transmission, as it can be run locally for specific use cases or orchestrated via GPU clusters from third-party providers to eliminate data transmission risks.
The company Perplexity, for example, offers a locally hosted version of DeepSeek-R1 on servers located in the U.S. and Europe, ensuring the safety of user data. This means that data is secure as long as it’s going to a locally hosted version of DeepSeek-R1, whether on the user’s machine or a GPU cluster in the Western region.
Chinese startup DeepSeek released an open-source model, DeepSeek-R1, that uses pure reinforcement learning to challenge the notion that only large-scale AI training can lead to high performance. Concerns emerged around the company’s privacy policy, which collects and stores user data in China and raises alarms about security and privacy implications. However, the open-source DeepSeek-R1 model itself does not involve data transmission and can be run locally or through secure third-party services, ensuring data safety. The next step is for users to consider the privacy implications and choose the most secure option for their needs.