The Chinese AI model DeepSeek-R1 has been making headlines worldwide in recent days. The project is rapidly gaining popularity and has already contributed to the decline of major U.S. tech stocks. However, some analysts point out that the model is subject to strict censorship and may be transmitting data to the Chinese government.

A study by PromptFoo, a startup backed by Andreessen Horowitz, found that DeepSeek-R1 ignores 85% of queries related to “sensitive” topics concerning China. These include questions about Chinese protests, the Tiananmen Square incident, and Taiwan’s status.

Analysts noted that instead of providing responses, DeepSeek frequently offers nationalist-leaning statements. They argue that the censorship system in the model has been implemented too bluntly, making its behavior predictable.

Additionally, researchers have identified vulnerabilities in DeepSeek that allow users to bypass restrictions. This suggests that experienced users could override censorship limitations. Moreover, these “loopholes” could expose the model to misuse for illicit purposes.

The growing popularity of DeepSeek, which outperforms competitors in terms of cost and efficiency, has raised concerns among U.S. tech giants. Recently, Microsoft and OpenAI launched an investigation into the Chinese startup.

This came after specialists detected suspicious activity related to potential unauthorized data extraction from OpenAI’s technology stack. A large-scale data download via API was recorded in late 2024.

Reminder: White House Director of AI and Cryptocurrency, David Oliver Sacks, recently commented on the launch of DeepSeek-R1.