The DeepSeek app and website are safe in the sense that they won’t infect your phone, but they aren’t private. DeepSeek’s privacy policy says it stores personal data in the People’s Republic of China, collects your prompts, uploaded files, photos and chat history, and uses that material to train its models unless you email them to opt out. The open-weight DeepSeek models are a different case: run on your own hardware, nothing reaches DeepSeek at all.
So the useful question is less “is DeepSeek safe” and more “what am I comfortable sending to it.”
What does the DeepSeek app collect? #
DeepSeek’s privacy policy (last updated February 10, 2026) lists these categories:
| What | Examples from the policy |
|---|---|
| What you type and upload | Text and voice input, prompts, uploaded files, photos, chat history |
| Device information | Device model, operating system, IP address, device identifiers |
| Usage | Features you use and actions you take in the app |
| Location | Approximate location based on your IP address |
The policy says DeepSeek won’t extract voiceprints or facial recognition data from the voice and photos you send. It also says your data is used for “training and improving our technology, such as our machine learning models.” You can opt out of training, but not with a toggle: the policy tells you to email [email protected]. You can read the full text in DeepSeek’s privacy policy.
Most of that list looks like any cloud chatbot’s. ChatGPT, Gemini and Claude also keep your chats on their servers and, depending on your settings, may use them for training. The difference is the next section.
Why does it matter that the data is stored in China? #
Where a company stores data decides which governments can demand it. DeepSeek is a Chinese company storing data in China, so Chinese law governs access. China’s 2017 National Intelligence Law requires organizations to support and cooperate with state intelligence work, and companies there have little room to refuse a request or tell users about it.
That doesn’t mean anyone is reading your recipe questions. It means you should treat anything you send as potentially available to a government you have no legal relationship with. For most casual use that’s an acceptable trade. It isn’t for work documents, client information, health details, or anything about people who didn’t agree to be discussed.
Who has banned DeepSeek? #
Governments have mostly restricted it on official devices rather than for the public:
- United States: the National Defense Authorization Act for Fiscal Year 2026 directs the removal of DeepSeek AI from Department of Defense and intelligence community devices, with narrow exceptions for research and security work.
- Australia: the Department of Home Affairs directed government agencies to remove DeepSeek products and block their use on government systems and devices.
Other agencies and regulators have taken their own steps, and the list keeps changing. If you use a work phone, check your employer’s policy before installing it. Many companies block DeepSeek along with other cloud chatbots that haven’t been approved.
Is the DeepSeek app itself a security risk? #
Installing it from the official App Store or Google Play listing is no riskier than installing any other big-name app. The risks are about data, not malware:
- Account data. Signing in links your chats to your email or phone number.
- Permissions. Voice input needs the microphone; photo questions need the camera or photo library. Grant them only while you use those features.
- Copycats. Search results fill with fake “DeepSeek” apps after every news cycle. Check the developer name before you install.
What’s the difference between the DeepSeek app and the DeepSeek models? #
This trips a lot of people up. DeepSeek does two separate things:
- Runs a chatbot service (the app and chat.deepseek.com). Your messages go to its servers. This is what the privacy policy covers.
- Publishes open-weight models. DeepSeek-R1, released on January 20, 2025, came out under the MIT License, so anyone can download and run the weights themselves. Run them on your own computer and no data goes to DeepSeek.
The catch for phones is size. The full R1 and V3 models are far too big for any phone or ordinary laptop, so local users run much smaller versions, and those are a different, weaker product from the chatbot in the app.
Bias is a separate question from privacy. A model trained in China may still reflect Chinese content rules on political topics even when it runs offline. We look at that for another Chinese model family in is Qwen safe to use.
How to use DeepSeek more safely #
If you like DeepSeek’s answers and want to keep using the app:
- Keep personal and work data out. No names, addresses, account numbers, client details or medical specifics. Our guide to keeping personal data out of AI chatbots shows how to strip a prompt.
- Opt out of training by emailing [email protected], as the policy describes. Keep a copy of the email.
- Use it logged out of anything else. Don’t sign in with the same account you use for work tools.
- Limit permissions to “while using the app” and deny contacts access.
- Delete old chats and, if you stop using it, your account.
- Don’t install it on a work phone unless your employer allows it.
A private alternative: run a model on your phone #
If what you want is a capable AI you can tell anything, the more private route is a model that runs entirely on your own phone. Nothing is uploaded, so there’s nothing to store, train on or hand over.
Personal LLM does this with open models such as Qwen 3.5, Gemma 4, Ministral 3 and GLM 4.6V Flash, or any GGUF model you add by URL. You download a model once from Hugging Face, and after that chats, photos and documents stay on the device and it works in airplane mode. There’s no account and the app has no server of its own. The free version shows ads through Google AdMob when you’re online; a one-time purchase removes them.
Be realistic about the trade: a 4B or 9B model on a phone won’t match DeepSeek’s full-size chatbot on hard reasoning. For drafting, summarizing, explaining and sensitive personal questions it’s often enough. To judge any AI app’s claims yourself, use our checklist for telling if an AI app is private.
Frequently asked questions #
Does DeepSeek send data to China? #
Yes. DeepSeek’s privacy policy says the personal data it collects, including your chats, uploaded files and device information, is stored in the People’s Republic of China. That applies to the app and website, not to DeepSeek’s open-weight models run on your own hardware.
Is DeepSeek safer than ChatGPT? #
Neither is private: both store your chats on their servers. ChatGPT is a US company with in-app controls for training and chat history. DeepSeek stores data in China under Chinese law, and opting out of training means emailing the company. For sensitive material, the difference matters more than the quality of the answers.
Can I use DeepSeek offline? #
Not the app. It needs a connection because the model runs on DeepSeek’s servers. You can run smaller open-weight DeepSeek models offline on a computer, and you can run other open models offline on a phone with an on-device AI app.
Is DeepSeek banned in the United States? #
Not for the general public. The 2026 defense authorization law directs its removal from Department of Defense and intelligence community devices, and other public bodies have restricted it on their own devices. Individuals can still download it.
Should I delete DeepSeek? #
If you’ve shared personal or work information in it, deleting those chats and your account is a reasonable step. If you only ask general questions, keeping it is a judgment call about whether you’re comfortable with your data being stored in China.