Is DeepSeek safe? What its policies say and how governments have responded

The short answer

There is no official safety rating for DeepSeek. What its own pages say: as of October 29, 2026, DeepSeek's privacy policy says it directly collects, processes and stores personal data in the People's Republic of China, and lists training its models among the uses of that data. Its terms say users can opt out of having chats used to improve the service by turning off a setting called Improve the model for everyone.

Governments have acted mainly on government use. Twelve U.S. states we have confirmed, the Defense Department under a 2025 law, and federal civilian agencies under an October 2026 Homeland Security order restrict DeepSeek on government devices, networks or contracts. Australia barred it from government systems in February 2025, and Italy's data protection authority ordered a limitation on processing Italian users' data in January 2025. The U.S. rules we found do not bar private individuals from using it on their own devices, though Virginia's order bars anyone from using it over state-owned wireless networks.

Who can use it

DeepSeek's terms of use, last updated March 27, 2026, say the service is primarily intended for adults. Users under 18, or under the minimum age required in their country, should review the terms with a guardian and use the service only with guardian consent. The terms do not set a fixed minimum age such as 13.

DeepSeek's privacy policy says the service is not aimed at children and that DeepSeek does not knowingly process children's personal data. It says that if DeepSeek learns it has collected such data, it will investigate and delete it where appropriate, and that concerns can be sent to privacy@deepseek.com.

DeepSeek's terms say the service is operated by Hangzhou DeepSeek Artificial Intelligence Co., Ltd., that the terms are governed by the laws of mainland China, and that disputes may be brought in a court with jurisdiction over the company's registered office.

What DeepSeek collects and where it stores it

DeepSeek's privacy policy, last updated December 22, 2025, lists account data, user input, information sent when contacting DeepSeek, device and network data, log data and location data. It says data collected automatically includes IP address, unique device identifiers and cookies, and that cookies and payment data do not apply to the DeepSeek app.

The policy says DeepSeek collects, processes and stores personal data directly in the People's Republic of China. It says DeepSeek may share data with its corporate group for purposes including storage, security, research and support, and with law enforcement agencies, public authorities and others when it believes in good faith that doing so is necessary.

The policy asks users not to provide sensitive personal data, such as health or religious information. It also says conversations shared by a public link may be picked up by web crawlers if the link is published.

Is DeepSeek trained on your chats?

DeepSeek's privacy policy lists training and improving its machine learning models among the uses of personal data. Its terms say DeepSeek may use inputs and outputs to improve its services only after they are securely encrypted and strictly de-identified, and that users who do not want this can turn off the Improve the model for everyone setting.

DeepSeek's model and algorithm disclosure page says pre-training uses publicly available internet data and licensed data, and that a small portion of the question-and-answer data used for later optimization training may be based on user input, after de-identification and anonymization. It says users have the right to opt out of AI training.

Deleting chats and your account

DeepSeek's privacy policy says users can copy or delete their chat history in settings, and that deleting an account is permanent. Its terms say DeepSeek may keep certain data after an account is deleted where the law requires it.

The privacy policy says DeepSeek keeps personal data for as long as an account is active, and may keep it longer to meet legal obligations, deal with policy violations, protect legitimate business interests or handle legal claims. It does not give fixed retention periods.

What its rules prohibit

DeepSeek's terms prohibit using the service to generate content that is hateful, harassing, discriminatory or violent, sexually explicit content, content that exploits minors, and instructions for crimes. They say outputs may contain errors and should not be treated as professional advice, and that outputs used for significant decisions about people need human review.

What the app store listings say

• Apple App Store (viewed October 29, 2026): DeepSeek - AI Assistant, from Hangzhou DeepSeek, is free and rated 13+. In privacy details the developer provides, which Apple says it has not verified, the listing says data linked to the user includes coarse location, contact info, user content, search history, identifiers, usage data and diagnostics.
• Google Play (viewed October 29, 2026): the listing, as displayed when we viewed it, showed 100 million+ downloads and a 12+ rating with parental guidance recommended; ratings shown can vary by region. Its data safety section says the app may share device or other IDs with third parties, collects location, personal info and other data types, encrypts data in transit, and lets users request deletion.

U.S. government restrictions

These rules cover government devices, networks and contracts, not private use. Our guide to DeepSeek bans by state lists each state action we have confirmed with its official source.

• States: the earliest state action we confirmed is a Texas governor's directive of January 31, 2025. New York, Virginia, Kansas and other states followed with orders, agency policies or, in Kansas, a statute. Sources for each state are in our guide to DeepSeek bans by state.
• Defense Department: the National Defense Authorization Act for Fiscal Year 2026, approved December 18, 2025, includes a prohibition on Defense Department use of covered AI, including DeepSeek, according to the Congressional Research Service, and a section on removing DeepSeek from intelligence community systems.
• Federal civilian agencies: a Homeland Security order published in the Federal Register on October 20, 2026 excludes DeepSeek products from civilian agency procurement and requires their removal from civilian agency IT, including third-party products built on DeepSeek.
• 21 state attorneys general, led by Montana's, urged Congress on March 6, 2025 to pass the No DeepSeek on Government Devices Act. That bill had not been enacted as of October 29, 2026.

Foreign regulators

• Italy: on January 30, 2025, Italy's data protection authority, the Garante, ordered an immediate limitation on the processing of personal data of users in Italy by Hangzhou DeepSeek and Beijing DeepSeek, and opened an investigation. Its decision says the companies told it they did not operate in Italy and that EU data protection law did not apply to them, which the Garante rejected. The decision lists concerns including the lack of an EU representative and data storage in China.
• Australia: the Department of Home Affairs published a protective security direction on February 4, 2025 requiring Australian Government entities to prevent the use or installation of DeepSeek products, applications and web services and to remove existing instances from government systems and devices.

Which laws may apply

No U.S. law we found bars individuals from using DeepSeek, and none of the laws below names it. Whether DeepSeek counts as an AI companion or a conversational AI service under a particular state law is a legal question. Some examples:

• Maine LD 1727, signed June 12, 2025 and in effect since September 24, 2025: a person may not use an AI chatbot in trade and commerce with a consumer in a way that could mislead a reasonable consumer into thinking they are dealing with a human, unless the consumer gets clear and conspicuous notice that they are not. A violation is a violation of the Maine Unfair Trade Practices Act.
• New York's AI companion law (General Business Law sections 1701 and 1702): operators of AI companions must make reasonable efforts to detect expressions of suicidal thoughts or self-harm and refer users to crisis services, and must tell users they are not talking to a human at the start of an interaction and at least every three hours during continuing ones. The law defines AI companions as systems designed to simulate a sustained relationship, and excludes tools primarily designed for efficiency; it has been in effect since November 5, 2025.
• Idaho's Conversational AI Safety Act (S 1297), signed March 31, 2026 and effective July 1, 2027: operators of conversational AI services open to the public must disclose that the service is AI when people could think they are talking to a human, adopt a suicide-response protocol with crisis referrals, and meet extra duties for minor account holders. The law excludes some tools, and the Attorney General enforces it.

We are not saying DeepSeek is covered by, or out of compliance with, any of these laws. It may have practices it has not published, and whether a law applies to it is a legal question this guide does not decide.