DeepSeek: Innovation or Privacy Risk?

“The primary concern is that the AI model may leak data to the Chinese government.”

With this statement, the Dutch Data Protection Authority (AP) issued a warning on February 3, 2025, regarding DeepSeek, a Chinese AI chatbot that gained global popularity in a short period. The chatbot was downloaded en masse, surpassing ChatGPT in usage, but simultaneously led to growing concerns over privacy and data security. What makes DeepSeek so controversial? What privacy risks does using DeepSeek’s models entail? And are there positive aspects to this AI model, or is it entirely negative?

What is DeepSeek?

DeepSeek is an AI development company, founded in May 2023 and based in Hangzhou. It focuses on the development of open-source Large Language Models (LLMs) and has since launched several generative AI models.

In late January 2025, DeepSeek gained worldwide fame with the release of DeepSeek-R1, an AI model that delivers performance comparable to OpenAI’s GPT-4o, but at a fraction of the cost. While competing models cost hundreds of millions of dollars to develop, R1 reportedly cost less than $6 million.

The rapid rise of DeepSeek has had a major impact:

• Massive adoption – Within days, the DeepSeek AI assistant became the most downloaded AI app in Apple’s App Store.
• Market effect – On January 27, 2025, the growing popularity of DeepSeek led to a decline in the share prices of American tech companies such as Nvidia.
• Versatile applications – DeepSeek offers various AI models and applications, including DeepSeek Coder (for programming tasks), DeepSeek-V2 (an earlier version of the model), and Janus-Pro-7B (a multimodal model for text and images).

Despite technological advancements, DeepSeek raises questions about data protection and the role of the Chinese government in data processing.

Privacy Risks and Data Security

The greatest risk of Chinese AI models lies in the processing and storage of user data. Because DeepSeek utilizes servers in China, there is a possibility that the Chinese government could gain access to users’ personal data.

Some key risks:

• Data storage – Inputted data, such as CVs or legal inquiries, may be stored on Chinese servers without clear guarantees regarding data protection.
• Unlawful transfer – The GDPR sets strict requirements for data transfers outside the EU. Without adequate safeguards, using DeepSeek may violate European privacy regulations because the data, as the AP also notes, is transferred to and stored in China. This aligns with concerns regarding a potential link between DeepSeek and China Mobile, a Chinese state-owned enterprise banned in the US due to ties with the Chinese government. Researchers discovered that the DeepSeek website contains code that potentially sends user data to China Mobile.
• Unclear policies – There is insufficient transparency regarding how DeepSeek handles personal data.

Government Measures and Bans

Several countries have now introduced restrictions on the use of DeepSeek:

• Italy – The Italian privacy watchdog has blocked DeepSeek due to a lack of transparency regarding data processing and potential GDPR violations.
• The Netherlands – Government employees are no longer permitted to use DeepSeek. The AP advises caution when entering personal data and warns of the risk of data transfer to Chinese servers.
• United States – Senator Josh Hawley has introduced a bill to ban Chinese AI models, including DeepSeek. This proposal blocks the import and use of Chinese AI technology. Violations could result in fines of $1 million for individuals, $100 million for companies, and prison sentences of up to 20 years.
• South Korea – The South Korean government has blocked DeepSeek for civil servants due to concerns over potential data leaks and Chinese state interference.
• Australia – Australia has banned DeepSeek for government use, citing “serious security concerns” and a lack of transparency regarding data management.

However, while DeepSeek is under pressure and bans are being implemented, Perplexity, for example, has chosen to integrate DeepSeek into its platform without exposing users to privacy risks. This is because Perplexity hosts the AI model on its own servers and does not share data with China.

The Positive Aspects of DeepSeek

The example of Perplexity demonstrates that DeepSeek does not only carry risks but also offers benefits. Unlike many closed AI models, DeepSeek distinguishes itself through its open-weight structure, which provides companies and researchers with greater flexibility and control.

A closed AI model, such as OpenAI’s ChatGPT, keeps model weights and training data secret. This means users have no insight into how the model works or what data it is based on. Open-weight models, such as DeepSeek-R1, offer more transparency: users can download the weights and fine-tune the model for specific applications. This allows organizations to deploy AI in a way that aligns with their own (sector-specific) preferences.

Another important feature is DeepSeek-R1’s ability to make the reasoning process transparent. This means users do not just receive an answer but can also see how the model arrived at a particular conclusion. This is especially valuable in sectors where transparency is crucial, such as the legal and financial sectors.

Although DeepSeek must censor certain topics within the official application in China, users can run the model locally or deploy it on an independent cloud server to maintain full control over the output, as seen in the Perplexity example.

Conclusion

The rise of Chinese AI models like those from DeepSeek demonstrates how rapidly the Chinese AI industry is developing and the far-reaching impact this has on the global technology sector. While powerful and innovative capabilities are offered, fundamental concerns regarding data processing and potential government interference remain.

Countries such as Italy, the Netherlands, the US, South Korea, and Australia have already taken measures to restrict the use of DeepSeek. The question is how other Western nations will respond. Will there be a broader ban, as previously attempted with TikTok? Or will the approach continue to vary by country?

Until there is more clarity on how Chinese AI companies handle data protection, caution remains advised when using Chinese AI models.