DeepSeek is a new AI platform that exists both as a freely accessible chatbot and as an open source model. Developed in China, DeepSeek is often seen as an alternative to well-known AI systems such as ChatGPT or Gemini.
We are currently reading a lot of good things about the new AI bot in the media. It is often reported that our data is much better protected at DeepSeek thanks to the open source model. It is also emphasized that the company has developed such a powerful model with significantly fewer resources and budget that it is almost comparable to ChatGPT.
But how secure is our data really at DeepSeek and what do we need to know about data protection? When it comes to choosing the right AI chatbot, such as ChatGPT, SwissGPT or others, we are usually faced with the choice between more data protection or a greater range of functions. While some users are primarily looking for powerful functions, others attach great importance to how their data is processed and stored.
Opinions differ, particularly with regard to DeepSeek: Some AI users see the new open source variant as a clear advantage for data protection, while others express concerns about the origin of the company and the handling of the data, as well as the training data on which the model is based.
This article examines both aspects and compares DeepSeek directly with other prominent AI chatbots such as ChatGPT, Gemini, Perplexity and others.
DeepSeek is not just DeepSeek – what variants are there?
DeepSeek as a chatbot from China – DeepSeek as a proprietary model
DeepSeek is a powerful AI-supported chatbot that is freely accessible to a broad user base. It is often mentioned as an alternative to ChatGPT or Gemini, as it is able to precisely understand and respond to complex language queries. What is particularly interesting is that DeepSeek was developed in China and therefore offers an independent alternative to the market-leading US AI solutions.
For many companies and individual users, DeepSeek is an attractive option as it allows them to use powerful voice AI without being tied to the US technology ecosystems. The impact of the Chinese origin on data protection and data security is often discussed.
DeepSeek stores user data on servers in China, which has led to concerns about data security and access by the Chinese government. The data collected is used to improve the performance of the model and optimize the user experience. Due to storage in China, there are concerns that international data protection regulations and security precautions may not be fully observed, which could be particularly problematic for users from countries with strict data protection laws. Experts and data protection authorities have pointed out these risks. The use of DeepSeek may therefore raise data protection concerns, especially in regions with high data protection requirements, such as the EU and the USA. This means that users may not be fully informed about how their data is processed and whether it complies with local data protection laws.
Another potential risk factor is access by the Chinese government, which could potentially access the stored user data due to Chinese data surveillance law. While DeepSeek may be taking measures to protect privacy and ensure data security, it remains unclear to what extent these security protocols are sufficient to meet international requirements. It is recommended that organizations using DeepSeek carefully review the provider’s privacy policies and security protocols, particularly with respect to data storage and access.
In summary, the public AI chatbot Deepseek stores all user data and chat conversations. It is not possible to delete this data.
DeepSeek as an open source model – what does that mean?
In addition to the cloud version, i.e. the public AI chatbot described above, DeepSeek also offers an open source version that can be hosted on its own servers. This enables companies and organizations to retain full control over their data and minimize security risks from external processing.
Opportunities and challenges of self-hosting:
Open source models such as DeepSeek’s offer a high level of transparency, as the source code is publicly accessible. This means that developers and organizations can see how the model works and how data is processed. A major advantage of an open source model is that it can be hosted locally. This means that user data is not sent to central servers, but is processed on the user’s own servers. This reduces the risk of sensitive data being passed on to third parties or accessed by external parties.
Local hosting gives companies and individuals complete control over the storage and processing of their data. This is particularly important for organizations that have to comply with strict data protection regulations, such as the EU’s GDPR (General Data Protection Regulation). However, organizations must ensure that they implement the open source model correctly to ensure the security and privacy of user data, as open source software can potentially have security vulnerabilities if not configured correctly.
The technical implementation therefore requires specialist knowledge and human resources to set up and maintain the infrastructure. There are also ongoing costs for hardware, energy and IT security, which can be higher than using a cloud-based solution. Software updates and regular optimizations must also be carried out independently in order to maintain the performance of the AI and minimize security risks.
It is also possible to rely on solutions such as SwissGPT from AlpineAI or PeakPrivacy, which already work with open source models and can integrate DeepSeek into their environment in a data protection-safe manner.
Comparison of AI models in terms of data protection
| AI model | Data protection | Storage of user data | Access to training data |
| DeepSeek Open Source | Depending on the operator | No central storage | Transparency through open source |
| DeepSeek Proprietary | Developer guidelines | Possible storage in the cloud | Restricted access |
| ChatGPT (OpenAI) | Proprietary, but with data protection measures | Storage for model improvement (depending on use) | Closed architecture |
| Perplexity AI | More data protection-friendly than OpenAI, no long-term storage | Anonymized processing | Model architecture remains secret |
| SwissGPT | Focus on European and Swiss data protection guidelines | Local storage possible | Closed architecture |
| Gemini (Google) | Data-driven use, often with user profile linking | Storage for the optimization of Google products | Not open for external developers |
Use and creation of training data
An essential aspect of data protection is the question of how AI models are trained. DeepSeek pursues different strategies depending on the model variant:
DeepSeek Open Source
Training can be carried out with your own local data sets, giving companies more control over the data used. This means that data protection requirements can also be implemented in a targeted manner.
DeepSeek Proprietary, publicly accessible chatbot
Here, training is carried out on large data sets, which generally come from publicly accessible sources. Since DeepSeek is less restrictive than Google (Gemini) or OpenAI (ChatGPT) with the training data, this could lead to concerns about the origin and fairness of the data.
Other AI chatbots
OpenAI and Google obtain most of their training data from the internet, without users or companies knowing exactly what data is used in the models. This leads to considerable data protection risks, as sensitive or personal data can be unintentionally included in the model. Google’s Gemini in particular is strongly linked to existing Google services and uses user data to continuously optimize its models, which is associated with considerable data protection concerns. ChatGPT temporarily stores user input and potentially uses it to further develop the model. Perplexity AI, on the other hand, takes a different approach: it relies on targeted, high-quality data sources.
SwissGPT, on the other hand, offers a more data protection-friendly alternative, as it adheres more closely to European and Swiss data protection regulations and offers the option of storing data locally without the need for cloud storage.
Functional differences between the AI models
The various AI models offer different functions and focuses, with DeepSeek standing out due to its wide range of options.
DeepSeek
DeepSeek combines classic AI functions such as text generation, question answering and translation with improved multimodal processing for complex tasks. In particular, it supports the processing of texts in English and Chinese. It also offers an open source version that allows developers to adapt the model to their own requirements. This represents a clear advantage over completely proprietary models.
ChatGPT (OpenAI)
ChatGPT is designed for a variety of use cases, including conversation, texting and programming support. It supports over 99 languages and, depending on the subscription model, offers extended functions such as real-time access to the Internet. In addition, it offers exclusive functions such as ChatGPT Canvas, the creation and use of divisible CustomGPTs and ChatGPT Projects, which enable a structured way of working with AI. However, OpenAI temporarily stores user input to improve the model.
Perplexity AI
This model is characterized by an AI-supported search function that aggregates and presents structured information from the Internet and provides a transparent source citation for each query.
SwissGPT: Developed by AlpineAI, SwissGPT focuses on secure and privacy-compliant AI applications with a strong focus on European and Swiss data protection guidelines. It allows companies to host the model locally, ensuring maximum control over data.
Gemini (Google)
Gemini is Google’s answer to the ongoing development of AI and offers multimodal processing of text, images and code. However, Gemini is deeply embedded in the Google ecosystem, which means that user data is potentially used to improve other Google services. The model has real-time access to the Internet, which makes it particularly powerful, but also raises major data protection concerns.
Overall, the functional scope of ChatGPT as of 01-2025 appears to be the most advanced. Especially “additional functions” such as CustomGPTs, projects or canvas are mainly available with ChatGPT.
Pricing of the AI models
The price models of the AI models vary considerably:
- DeepSeek: Currently offers a free version for open source use.
- ChatGPT: OpenAI offers several price levels. The free version has limited features, while ChatGPT Plus offers advanced features for about 20 USD/month. Enterprise tariffs with additional security and data protection features are also available.
- Perplexity AI: The standard version is available for free, while Perplexity Pro offers an advanced AI-powered search with more features for 20 USD/month.
- SwissGPT: Pricing on request, as the model is primarily aimed at companies and can be hosted individually.
- Gemini: Google offers Gemini Advanced for CHF 17/month, with extended access to AI models and additional Google services.
Conclusion: Is DeepSeek a safe alternative?
As an open source model, DeepSeek offers an interesting alternative for companies that prioritize data protection and want to retain full control over their data. However, most users currently use the public DeepSeek, which does not offer comprehensive data protection mechanisms. The risk here is that user input is stored centrally and used for model improvements without users having full control over it.
At the same time, the proprietary version harbors potential data protection risks, especially if data is stored not only for direct processing, but also for future training of the model. This is a key difference to SwissGPT, which is more protected by a more regulated Swiss and European data protection policy and offers companies more options for local data storage.
A direct comparison with ChatGPT and Gemini shows that these models also harbor considerable data protection risks. Google’s Gemini is closely linked to Google services and uses user data to further optimize the model. This can lead to massive profiling, as Google already has a broad base of data about its users. OpenAI’s ChatGPT stores user input temporarily and can use it for the further development of the model, which can be problematic, especially in sensitive corporate environments.
Companies and users should therefore carefully consider which model best suits their requirements – especially if data protection plays a key role. If you want to retain the greatest possible control over your data, you should consider DeepSeek in a self-hosted open source version or SwissGPT.
A direct comparison with ChatGPT, Gemini and SwissGPT shows that DeepSeek has advantages and disadvantages depending on the type of application. While OpenAI and Google implement non-transparent data protection measures, they also offer a rather wide range of functions. SwissGPT is particularly suitable for companies with EU and Swiss data protection requirements.
DeepSeek, ChatGPT, Perplexity, SwissGPT and co. in comparison
Below is a comprehensive comparison of the best-known AI assistants with a special focus on the use of chat data, user information and training data.
See for yourself how safe DeepSeek is compared to the other AI chatbots.
Further questions
Has this article given you food for thought and do you have any further questions? Or are you looking for general support with the use of AI, ChatGPT, Deepseek and chatbots?
I am always happy to receive your messages, preferably by WhatsApp message or e-mail.
This article is also available as a podcast episode
Attention! The podcast was created entirely by my AI-Assistant based on my contribution – no guarantee for incorrect content.
