DeepSeek made global headlines in early 2025 by claiming to match leading AI models at a fraction of the training cost, but the safety and privacy questions that followed have proven more significant, and better documented, than the initial cost story. This guide explains what those concerns actually are, which governments have restricted it and why, and how to use DeepSeek’s tools more safely if you choose to.
Photo by Towfiqu barbhuiya on Unsplash
What Is DeepSeek?
DeepSeek is a Chinese AI company whose chatbot works similarly to ChatGPT, Claude, or Gemini: users can ask questions, generate code, draft text, and analyse data through a conversational interface. Its distinguishing claim was training efficiency, reportedly developing its V3 model for a fraction of what Western competitors spent on comparable models, which raised genuine questions in the industry about whether massive AI infrastructure investment was strictly necessary. That efficiency story, however, is separate from the data handling and security concerns that have driven most of the regulatory action against it.
The Core Data Privacy Concern
DeepSeek’s own privacy policy states plainly that personal data collected from users is processed and stored in the People’s Republic of China. This matters because Chinese law, specifically the 2017 National Intelligence Law, requires organisations and citizens to support and cooperate with state intelligence work when compelled. This creates a legal framework that international users and businesses cannot contract around, regardless of DeepSeek’s own stated intentions regarding user data. South Korea’s data protection regulator additionally found that DeepSeek had transferred user prompts to other Chinese companies without user consent, adding a documented data-sharing concern on top of the jurisdictional issue.
Documented Security Weaknesses
Beyond data jurisdiction, independent security testing has raised separate concerns. Testing by Cisco researchers found that DeepSeek’s R1 model failed to block any of the jailbreak attempts used in their evaluation, meaning the model’s built-in safety guardrails were bypassed in every test case attempted. A separate testing firm reported DeepSeek failing over half of its jailbreak tests. A database exposure incident was also documented, reportedly affecting a significant number of records. Together, these findings suggest weaker safety engineering compared with several Western competitors, independent of the data jurisdiction question.
Which Governments Have Restricted DeepSeek
| Country/Region | Action Taken |
|---|---|
| Italy | Blocked from app stores by the Garante over GDPR data-handling concerns |
| Australia | Banned from all government devices and systems |
| Taiwan | Prohibited across public sector organisations, state-owned enterprises, and public schools |
| South Korea | Temporarily suspended downloads nationwide; multiple ministries banned official use |
| United States | Restricted on government devices in several agencies and states; a House committee report labelled it a national security concern |
Most of these actions specifically target the hosted consumer app and its data handling practices, rather than the underlying open-weight models as a purely mathematical object, a distinction that matters for the self-hosting option covered below.
Does Self-Hosting Solve the Problem?
DeepSeek releases open-weight versions of its models that can be downloaded and run on your own hardware or a Western cloud provider rather than through DeepSeek’s hosted app. Doing this removes the cross-border data-flow concern entirely, since your prompts never reach DeepSeek’s servers in China at all. It does not, however, resolve the model’s own underlying safety weaknesses, such as the documented jailbreak vulnerability, which are properties of the model itself rather than the hosting arrangement.
Practical Guidance for Different Use Cases
- Casual, non-sensitive personal use: Risk is comparable to using any free foreign chatbot, though the documented data jurisdiction issue still applies to the hosted app specifically
- Business or professional use involving proprietary or client data: The hosted app is not appropriate given the demonstrated inability to contractually guarantee data protection under Chinese jurisdiction
- Regulated data (health, legal, financial, government): Multiple regulators have explicitly stated the hosted service should not be used for this category of data
- Developers wanting the cost efficiency without the data concern: Self-hosting the open-weight models on Western infrastructure addresses the jurisdictional issue specifically, though not the model’s own safety weaknesses
If you’re comparing DeepSeek against mainstream Western alternatives, our Claude vs ChatGPT vs Gemini comparison covers how those established assistants stack up.
Frequently Asked Questions
Is DeepSeek’s data really sent to China?
Yes, according to DeepSeek’s own privacy policy, which states that personal data is directly collected, processed, and stored in the People’s Republic of China.
Is DeepSeek a scam?
No, it is a legitimate AI company with genuinely functional models. The concerns are specifically about data jurisdiction, documented security weaknesses, and government restrictions, not fraudulent operation.
Can I use DeepSeek safely for casual coding or study?
Many assessments suggest lower relative risk for casual, non-sensitive use, provided you avoid entering personal, financial, or proprietary information into the hosted app specifically.
Does self-hosting DeepSeek eliminate all risk?
It removes the China data-jurisdiction concern specifically, but documented model-level safety weaknesses, such as jailbreak vulnerability, remain properties of the model regardless of where it’s hosted.
Conclusion
DeepSeek’s safety concerns are documented and substantial rather than speculative: a privacy policy that plainly states data is stored under Chinese jurisdiction, independent testing showing weak resistance to jailbreak attempts, and restrictions from multiple governments citing national security and data protection reasons. For casual personal use the risk profile is comparable to other free foreign AI tools, but for business, professional, or regulated data, the documented concerns are serious enough that multiple regulators have explicitly advised against using the hosted service.
