DeepSeek Asks Staff Not To Leave China, Takes Away Passports: Report

The passports of multiple employees of Chinese AI startup DeepSeek have been confiscated, preventing them from travelling outside the country, according to a new report. The government has reportedly tightened regulations on the company to prevent sensitive information from being leaked and key employees from deserting it.

DeepSeek's parent company, High-Flyer, a hedge fund, is holding onto the passports of key employees - especially those in research and development - to ensure they cannot leave the country without approval, The Information reported.

The report further claimed that the Chinese government now plays a direct role in deciding who can invest in DeepSeek, which gained global attention in January with the launch and meteoric rise of its open-source "reasoning" model, R1.

Chinese President Xi Jinping's administration has even labelled DeepSeek a "national treasure".

DeepSeek's AI technology has faced scrutiny for following China's censorship rules. Earlier, The Guardian reported that DeepSeek's chatbot avoids discussing sensitive political topics such as the Tiananmen Square massacre, the Hong Kong Umbrella Revolution, and other government-sensitive issues.

The restrictions come amid heightened AI competition between China and the US. Reports suggest that Beijing recently advised AI researchers and business leaders to avoid travelling to the US due to concerns about trade secret leaks.

Kylie Robison, a senior AI reporter at The Verge, cited three unnamed sources who claimed the purpose of these travel bans was to prevent the disclosure of confidential data, including trade or even state secrets.

Founded in May 2023 by Liang Wenfeng in Hangzhou, DeepSeek quickly became a significant player in the AI industry. The company first entered the AI space with DeepSeek Coder in November 2023, followed by DeepSeek LLM and DeepSeek V-2 in May 2024. Its most recent models, R1 and V-3, have drawn global attention for their impressive capabilities at a surprisingly low cost.

DeepSeek's AI model, DeepSeek-V3, reportedly costed under $6 million (Rs 51 crore) to develop - significantly less than what US tech giants like Microsoft and OpenAI spend. According to its creators, DeepSeek-V3 "tops the leaderboard among open-source models" and rivals leading closed-source AI models in math, coding, and natural language reasoning.

India's finance ministry recently warned employees against using AI tools like ChatGPT and DeepSeek on government devices, citing risks to official data privacy. The US has also imposed export bans on high-end Nvidia A100 chips, limiting DeepSeek's access to critical hardware. The European Union is also considering tighter regulations on foreign AI applications to safeguard national security and user data.