also searched as: is deepseek safe · is deepseek safe to use · is deepseek private · is deepseek secure · how safe is deepseek · is deepseek safe to download · is it safe to run deepseek locally
The DeepSeek app is safe only for chats you would not mind a stranger reading, because its own privacy policy says your data is stored in China. Running the open DeepSeek models yourself is a different risk: your prompts never reach DeepSeek, but the model's weak spots come with it.
Use it if you want a free, capable chatbot for everyday questions and you switch training off first.
Be careful if you build agents or ship code with it, since US government testers found it far easier to jailbreak than US models.
Skip it if you handle client, patient or employer data, or use a government device.
sources checked 2026-10-13 · by dan scutaru
| what we checked | what we found |
|---|---|
| Owner and data location | China, by its own policy Hangzhou DeepSeek says it collects, processes and stores your personal data in the People's Republic of China (policy dated February 10, 2026). source |
| Training on your chats | On until you switch it off The terms say inputs may be used to improve the service unless you turn off Improve the model for everyone. source |
| What it collects | What you type, plus device data Text and voice input, files, photos, chat history, device identifiers, IP address and approximate location; it says it does not sell data. source |
| Wiz, January 2025 | Database left open Wiz found an unauthenticated DeepSeek database with over a million log lines, including chat history and API secrets; DeepSeek closed it after the report. source |
| NowSecure, February 2025 | Weak encryption in the iOS app The iPhone app sent some device data unencrypted and used 3DES with a hardcoded key; we found no published retest. source |
| NIST CAISI, September 2025 | 94% of malicious requests answered With a common jailbreak, DeepSeek R1-0528 complied with 94% of overtly malicious requests, against 8% for US models. source |
| Regulators | Italy ordered a stop Italy's data protection authority ordered DeepSeek to stop processing Italians' data on January 30, 2025, citing storage in China among its reasons. source |
| Age limits | 13+, guardian under 18 The terms require a guardian's consent under 18; the US App Store rates it 13+. source |
Yes. The app is run by Hangzhou DeepSeek, its terms fall under mainland Chinese law with disputes heard in Hangzhou, and the policy allows sharing data with authorities on government request. In June 2025, Berlin's data protection commissioner called the transfers to China unlawful, citing Chinese authorities' far-reaching access to personal data. Articles from early 2025 still say the policy collects keystroke patterns; the current version does not list them.
Run locally, DeepSeek keeps your prompts on your machine. Ollama's documentation says it does not see prompts from local models, and Amazon says prompts sent to DeepSeek on Bedrock are not shared with model providers. The model's behaviour stays, though. CAISI found agents built on R1-0528 were 12 times likelier to follow malicious instructions than US models, and CrowdStrike found DeepSeek-R1 wrote severely flawed code 27.2% of the time when a prompt mentioned Tibet, against about 19% without.
Not for personal use. New York barred it from state-managed devices and networks on February 10, 2025, and a Senate bill from June 2025 that would push federal agencies to drop AI from foreign adversaries has not passed that we found. Abroad, the app is still missing from Apple's Italian store, returned to South Korea in April 2025, and remains listed in Germany despite Berlin's request to block it.
Hangzhou DeepSeek Artificial Intelligence Co., Ltd., a Chinese company, per its App Store listing and privacy policy. New York's state IT office describes it as run by High-Flyer, a Chinese trading firm.
Being in the US does not change where your data goes: the policy stores it in China for every user. Personal use is legal; the bans cover government devices.
Yes. Its terms describe filtering for legal compliance, and CAISI found DeepSeek models repeated four times as many misleading CCP narratives as US models.
Yes, it is a real company with a real app and published models. The official site is deepseek.com; a look-alike under another name is not DeepSeek.