If you type a question into the DeepSeek chatbot, where does that text actually go, who can read it, and can you stop it being used to train future models? Those are the three questions most people mean when they ask about DeepSeek privacy, and the honest answers are more nuanced than either the “it’s a Chinese data vacuum” headlines or the “it’s just another AI app” defences suggest. This guide walks through what DeepSeek’s own policy says it collects, where that data is processed, what independent researchers actually observed, the regulatory actions against the company since January 2025, and the practical steps that reduce your exposure if you still want to use the product. You’ll leave knowing exactly which trade-offs you’re making.

The short answer on DeepSeek privacy

DeepSeek is operated by Hangzhou DeepSeek Artificial Intelligence Co., Ltd. The Services are provided and controlled by Hangzhou DeepSeek Artificial Intelligence Co., Ltd., with its registered address in China, which makes Chinese law the primary jurisdiction over the data you send. DeepSeek’s privacy policy acknowledges that its servers are located in the People’s Republic of China, and that when you access the services, your personal data may be processed and stored on those servers.

That single fact — data stored in China, under Chinese law — is what drives almost every regulatory action and every enterprise ban you have read about. Everything else in this guide is detail around that core trade-off.

What data DeepSeek actually collects

The current privacy policy, last updated February 10, 2026, is unusually broad. Independent reviews of its text identify three buckets:

1. Data you provide

• Profile details: username, email, phone number, password, date of birth.
• Chat inputs, uploaded files, feedback.
• Payment information if you use the paid API. When you use paid services in the open platform, DeepSeek collects payment order and transaction personal data to provide order placement, payment, customer service, and after-sales support.

2. Data collected automatically

Device identifiers, IP addresses, cookies, and even keystroke patterns are collected during interactions with the platform. The keystroke-pattern clause is the one that drew the most attention when the app surged in January 2025 — it’s the kind of behavioural-biometric signal a VPN cannot mask.

3. Data from third parties

DeepSeek may receive personal data from sources such as log-in and sign-up via third-party services; if you sign in using Apple or Google, it collects data such as an access token from that service. The policy also allows sharing with its corporate group, and with advertising and analytics partners.

How much of that is real, versus just policy boilerplate?

This is where things get interesting. Privado AI’s technical analysis of the Android app found a mismatch between policy and practice: the privacy policy is broadly written and covers all possible data collection, including sensitive data types like keystrokes; in reality, DeepSeek actually collects less data than is declared, however, there are clear data flows to China. Their scan also observed data flows to China both to DeepSeek’s own servers and third-party SDKs integrated into the application, though concerns around keystrokes were not found in their tests. In plain English: the policy grants more latitude than the app currently uses, but the China-bound traffic is real.

Where your conversations go

Every message you send to the DeepSeek chat or DeepSeek app is processed on servers in China. That has two concrete implications:

1. Chinese law applies to the stored data. Law-enforcement access under legal process is possible, and companies operating in China can be compelled to provide user data on request.
2. GDPR routes are limited. DeepSeek’s engagement in Europe has been reluctant and reactive; the company initially claimed EU law did not apply to its operations — a position rejected by every DPA that considered it — and an EU representative was appointed only after Greek enforcement action compelled it.

The API is a different privacy story

Almost every privacy critique of DeepSeek is about the consumer app and web chat. The developer API has a different architecture that matters here.

DeepSeek’s current generation is DeepSeek V4, released April 24, 2026, shipped as two open-weight MoE models under the MIT license: deepseek-v4-pro (1.6T total / 49B active) and deepseek-v4-flash (284B / 13B active). Chat requests hit POST /chat/completions, the OpenAI-compatible endpoint at https://api.deepseek.com. Legacy model IDs deepseek-chat and deepseek-reasoner still work but route to deepseek-v4-flash and retire on 2026-07-24 at 15:59 UTC.

Crucially, the API is stateless: DeepSeek does not remember prior turns on its side — your client resends the conversation history with every request. That changes the privacy calculus: the web app stores session history server-side, but an API integration you build yourself controls what gets sent and what gets logged.

from openai import OpenAI
client = OpenAI(base_url="https://api.deepseek.com", api_key="...")
resp = client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=[{"role": "user", "content": "Summarise this note."}],
)

The API is OpenAI-compatible (and, since V4, also Anthropic-compatible) against the same base URL, so privacy-conscious teams can route through logging proxies, redact personal data before it leaves the network, and pin which fields ever reach the provider. See the DeepSeek API best practices for concrete patterns.

Training data: can you opt out?

Yes, with caveats. DeepSeek’s current privacy policy gives EU users the right to opt out of having their personal data used to train the models or optimise the technology, and the Terms of Use reference a setting to turn off “Improve the model for everyone”; it remains to be seen how DPAs around Europe can verify that such requests have been effectively implemented.

The opt-out exists inside the account settings of the web chat and app. If you are outside the EU, the availability and wording of that toggle can differ — check the settings pane after logging in.

Regulatory status: a moving target

DeepSeek has attracted the widest regulatory response of any consumer AI product to date. Italy’s Garante imposed a ban within 72 hours of launch, investigations followed in 13 European jurisdictions, the European Data Protection Board created a dedicated AI Enforcement Task Force, and government device bans proliferated from Washington to Canberra.

On Italy specifically, the nuance matters. The Garante’s order of January 30, 2025, targeted only the distribution of DeepSeek’s mobile app via official channels — DeepSeek was removed from the Italian Apple App Store and Google Play Store; the order did not target already-installed apps or the web version. So in practice, the “ban” is a store takedown plus a processing restriction, not an ISP-level block.

For a country-by-country snapshot that we keep current, see DeepSeek availability by country and the running coverage on DeepSeek US restrictions.

How DeepSeek compares to Western alternatives on data handling

It would be misleading to say DeepSeek collects uniquely unusual data. Western chatbots also log conversations, device data and usage patterns — see DeepSeek vs ChatGPT for a direct feature comparison. Two things set DeepSeek apart:

• Jurisdiction of storage. DeepSeek is legally required to comply with the Chinese government’s demands for data access and content control, with no legal recourse to resist; Western companies can challenge such requests in independent courts and can refuse requests that violate laws like GDPR.
• Advertising and analytics sharing. DeepSeek’s privacy policy allows sharing data with advertising or analytics partners, and covers personal information collected via cookies, web beacons and pixel tags, including chat history, device model, IP address, keystroke patterns, OS, payment information, and system language. Compare with OpenAI’s policy, which is more restrictive on marketing-purpose sharing.

Five ways to reduce your DeepSeek privacy exposure

1. Run the weights locally. V4-Pro and V4-Flash are MIT-licensed open-weight releases. If you have the hardware (or rent GPUs), see install DeepSeek locally or running DeepSeek on Ollama. No chats leave your network.
2. Use the API, not the app, for sensitive workflows. The stateless API lets you control logging, redact PII before sending, and pin traffic to your backend. Start with the DeepSeek API getting started guide.
3. Flip the training opt-out. In account settings, turn off “Improve the model for everyone” (exact label varies by region).
4. Avoid sensitive inputs entirely. Client records, health data, regulated secrets, and anything covered by NDA should not be pasted into any consumer chatbot — DeepSeek included. Review what your organisation allows with a pragmatic is DeepSeek safe walkthrough.
5. Delete your account if you stop using it. Walk through the steps in delete DeepSeek account; retention periods only start running once the account is gone.

Who should and shouldn’t use DeepSeek

Reasonable fit

• Hobby use, casual writing help, code snippets with no proprietary context.
• Self-hosted deployments of the open weights for teams that want the capability without the data flow.
• API-only integrations where you control what crosses the wire and you’ve reviewed the contract terms.

Avoid

• Regulated industries (healthcare, finance, legal) using the consumer app without a bespoke deployment.
• Government, defence, and critical-infrastructure workloads — multiple agencies have already banned it on official devices.
• Anything involving EU personal data under GDPR if you cannot demonstrate a lawful transfer mechanism.

For a roundup of privacy-first options, see our DeepSeek alternatives shortlist and the broader set of DeepSeek beginner guides.

Bottom line

DeepSeek privacy is a classic Pareto trade-off. You get frontier-tier capability at cost-efficient pricing in exchange for sending your inputs to servers in China, under a policy that permits wide data collection and advertising-partner sharing. That trade is acceptable for some workloads and unacceptable for others. If you must use the hosted service, flip the training opt-out, avoid sensitive content, and prefer the API over the app. If the jurisdiction is the blocker, the open weights are there — that is genuinely the exit hatch, and one reason DeepSeek’s regulatory story has not ended with an outright global shutdown.

Last verified: 2026-04-24. DeepSeek AI Guide is an independent resource and is not affiliated with DeepSeek or its parent company. Model IDs, pricing and API behaviour change; check the official DeepSeek documentation and pricing page before committing to a production decision.