Warning: DeepSeek products pose a risk to university environments

Masaryk University's cybersecurity team warns of the risks associated with using DeepSeek's AI tools. Based on the warning NÚKIB, we do not recommend using DeepSeek's products - either for work or personal use. Entering information into language models (e.g., AI chatbot) can seriously compromise data in a university environment.

15 Jul 2025 Warnings

No description

What's going on?

The NÚKIB warning is precautionary in nature - it highlights the long-term risks associated with DeepSeek products, not a specific incident. Below you will find a summary of the key findings, see the official warning for details [1].

Fingerprinting is the detailed tracking of technical parameters of a device (e.g. phone model, system version, language, time zone) that can lead to the unambiguous identification of a specific user.

Notice

The NÚKIB warning does not apply to DeepSeek's open-source language models when they are run purely locally without the ability to communicate with the company's servers. In practice, however, even a local model can send HTTP requests and thus pose a risk of data leakage unless its external communication is actively blocked (e.g. by network rules or DNS filters).

These concerns are not just theoretical...

DeepSeek and security? The problem is recurring. In January 2025, DeepSeek faced an attack in which attackers gained access to millions of user records. This included their conversation history, access data, API keys, etc. The company suspended new registrations, but it soon became clear that the data was freely available on the Internet, without any security [3].

In late May 2025, the Czech government officially named China as responsible for the cyber espionage campaign against the Ministry of Foreign Affairs.​The attackers were found to have gained access to thousands of emails, including those from the Czech EU presidency. This clearly shows how real and current the threat from Chinese state structures can be - and why the use of tools subject to their influence is a major security risk [4].

What to do?​​

Not only for the reasons given above, we stress that it is preferable to avoid their use. It is important to be aware of the consequences and we therefore give some specific situations that may arise when using them:

Conclusion

The combination of Chinese legislation, data collection and transmission, and technical vulnerabilities point to a high level of risk. Links to sanctioned entities are another problem. Instead, Microsoft Copilot can be used, which is available to all MU employees and students under their M365 account [5] (data processing is contracted in this case).

You can always find everything important about cybersecurity at Masaryk University at https://security.muni.cz/en.

You are running an old browser version. We recommend updating your browser to its latest version.

More info