How Does Deepseek’s Privacy Policy Compare To Other Ai Companies

Updated January 27, 2026: Added Microsoft Copilot privacy analysis.

This article is going to explore how DeepSeek’s privacy policy compares to the other ai companies’ privacy policies.

Companies who offer large-language models (like Open Ai, Claude, Google) all publish a privacy policy.

Privacy policies are boring, but they’re important because they state how your data can and can not be used.

When you interact with either a U.S or E.U. Ai company, you (the user) benefit from a privacy policy which protects your individual rights to a decent degree.

When you’re interacting with a China-based Ai company, like DeepSeek, the privacy policy is quite different, and the individual’s rights are not very well protected.

DeepSeek’s Privacy Policy Compared to Other Ai Companies

 

Let’s compare DeepSeek’s privacy policy against the major U.S. and E.U. Ai companies (and their Ai model names).

• • OpenAi (they make ChatGPT)
  • Anthropic (they make Claude)
  • Google (they make Gemini)
  • Meta (they make Llama)
  • Mistral (they make Mixtral)
  • Microsoft (they make Copilot)

When users interact with any of these bots through their official websites, (chatgpt.com, gemini.google.com , claude.ai, etc), that Ai company stores your chat & they can use it to train new Ai models, unless you explicitly opt-out (which they all allow the user to do).

If you want more-favorable privacy policy terms when you use an Ai model, then consider using those same exact Ai models through their API access, or through poe.com (which uses API access to connect).

Click this article to learn “What is Poe Ai“.

---

Difference between ‘Direct Use’ VS ‘API Access’

When you go directly to ChatGPT.com or download the DeepSeek app, you’re using the “consumer” version of that Ai.

Generally, consumer-versions of Ai models all ‘allow training on customer chats’, though it can generally be disabled on most company’s services.

However, you can also choose to use these exact-same Ai models through Poe, and when you do that, then you’re going through what’s called “API access”, which has a totally different (and better) privacy policy for the user.

---

DeepSeek’s Privacy Policy

DeepSeek became publicly available in January 2025 and it claims equal ability to other cutting-edge Ai models, like Chat GPT.

Direct Use:

• Your data gets used for training — no opt-out available
• Everything is stored in China, indefinitely
• Government access is required by law (China’s National Intelligence Law)
• There have been documented security vulnerabilities (1 million user records leaked in 2025)

API Access:

• Same policies as with direct (all training, no opt-out)
• Data still goes to China, forever
• Still used for training
• The only exception: if you self-host the open-source version, then you have full privacy control, since it’s not sending your info to an off-site server.

DeepSeek gets a 2 out of 10 on privacy, which is pretty bad.

OpenAI (GPT-4, GPT-5, ChatGPT) Privacy Policy

OpenAi makes the most popular Ai model in the world, which is Chat GPT.

Direct Use:

• They are allowed to train on your data by default
• User CAN opt out in settings
• Data is Stored in USA

API Access (including through Poe):

• Data and chats are NOT used for training
• All data and chats Deleted after 30 days
• Zero data retention is available for enterprise

Open Ai’s privacy policy gets an 8 out of 10, on that score card premise.

Anthropic (Claude) Privacy Policy

Anthropic makes the Ai model called Claude. In my opinion, Claude is the current best-all-around Ai model.

Direct Use (Claude.ai):

• As of September 2025, they changed their policy — may train on data now
• But you can opt out
• Stored in USA

API Access (including through Poe):

• NOT used for training
• Deleted after only 7 days — that’s the shortest retention of anyone
• Your identity isn’t even shared with them

Anthropic’s privacy policy gets a 9 out of 10.

Google (Gemini) Privacy Policy

Google makes Gemini and they make it available in as many places as possible, so it’s likely you’ve used Gemini without even trying to.

Direct Use:

• Used for training by default
• Humans may actually review your chats (anonymized, but still)
• EU/UK users get GDPR protections

API Access (Paid, including Poe):

• NOT used for training
• Deleted after 55 days

Free API Tier:

• May train on your data

I’d give Google’s privacy policy a 7 out of 10.

Meta (Llama) Privacy Policy

Llama is the Ai model from Meta (Facebook), and it’s powering basically anything that Meta makes (Instagram included).

Direct Use (Meta AI on Facebook/Instagram):

• May train on your data
• Linked to your Meta account

Through Poe or API:

• Meta never even sees your data
• The host (like Poe) controls everything
• Open source model

Surprisingly good privacy policy from Meta, if you use it through API or Poe. 9 out of 10.

Mistral Privacy Policy

Mistral is the European counterpart to OpenAi and Anthropic, and they make a variety of Ai models for the general public to use.

Free Tier:

• May use for training
• But you can opt out
• EU-based company

Paid API:

• NOT used for training
• Zero data retention available
• Can self-host in EU

Another 9 out of 10. These European companies take privacy seriously.

Microsoft Copilot Privacy Policy

[THIS IS AN UPDATED SECTION – ADDED JAN 27, 2026]

Microsoft Copilot is powered by multiple LLMs, primarily GPT-5 (as of mid-2025), with smart routing to GPT-4o, Claude, Gemini, and Microsoft’s own Phi-4 depending on the task. As of January 2026, Anthropic also operates as a Microsoft subprocessor for Copilot services.

Direct Use (Consumer Copilot):

• May use de-identified data for training by default
• User CAN opt out in Microsoft Account settings
• Stored in USA/EU (within Microsoft 365 boundary)

Enterprise Use (Microsoft 365 Copilot):

• Data NOT used for training foundation models
• Default retention: 30 days (admin-configurable via Microsoft Purview)
• Full compliance: GDPR, HIPAA, ISO 27001, ISO 42001
• Users can manually delete activity history anytime

Microsoft Copilot’s privacy policy gets an 8 out of 10.

---

Ai Privacy Policy Comparison

Provider	API Training?	How Long Kept?	Where Stored?	Privacy Score
DeepSeek	❌ Yes	Forever	🇨🇳 China	2/10
OpenAI	✅ No	30 days	🇺🇸 USA	8/10
Microsoft Copilot	✅ No*	30 days**	🇺🇸 USA / 🇪🇺 EU	8/10
Anthropic	✅ No	7 days	🇺🇸 USA	9/10
Google	✅ No*	55 days	🇺🇸 USA	7/10
Meta	✅ N/A	Host controls	Host	9/10
Mistral	✅ No	30 days	🇪🇺 EU	9/10

*M365 Copilot / Paid API only. Consumer/free tiers may train.
**M365 Copilot default; admin-configurable via Microsoft Purview.

---

How This Article Was Created

I made this article using the Script-Bot-Creator tool on Poe, which lets you do research, build things, generate software, and more, just through simple text conversation.

You just describe what you want, and it generates the output after doing all of the work; I’ve used Script Bot Creator to create reports and even video games.

If you want to see the whole conversation where we built this article and did all the research, check out the Script-Bot Creator chat on Poe, here.

---

What This Means for Poe Users

If you’re using AI through Poe, here’s the easiest summary:

For more on how Poe handles different AI tools, check out my articles on the best AI video generators in Poe and what is the best AI image generator — understanding these tools can help you make informed choices about which models to use for different tasks.

---

This article and the interactive comparison chart were created using Script-Bot-Creator on Poe.

The research, writing, and HTML visualization were all built through conversation with Claude.

All ideas and editing and spelling mistakes are Carleton’s