Someone on your team heard DeepSeek is free and just as smart as ChatGPT. They want to use it for client emails, proposals and spreadsheets.
Should you let them?
For anything with business or client information in it: no, not the DeepSeek app or website. The model itself isn't the main problem. Where your typing ends up is.
The quick answer
- Don't put client, employee or financial information into the DeepSeek app or chat website. Its privacy policy says it stores what you send on servers in China.
- For business work, use an AI tool on a business plan with a written promise not to train on your data. ChatGPT Business or Enterprise, Microsoft 365 Copilot Chat with a work account, or Gemini in Google Workspace.
- If you really want DeepSeek's model, use it through a U.S. cloud provider (Amazon Bedrock or Azure AI Foundry) or run it on your own hardware. That's a job for whoever runs your IT, not a browser tab.
- On a free consumer plan of anything, turn off training and keep client details out.
- Write it down. One line in your staff AI policy saying which tools are approved.
How they compare for a small business
This is the part that matters for most offices: where the data goes and who controls it.
Question | ChatGPT (business plans) | DeepSeek app and website |
|---|---|---|
Where your data is stored | OpenAI's servers; Enterprise and Edu can pick a storage region | Servers in China, per DeepSeek's privacy policy |
Trains on your chats? | Not by default on Business, Enterprise and Edu | Listed in the privacy policy as a right that may be available, depending on where you live |
Admin controls for a team | Yes: an admin console to add and remove users | No business workspace to manage |
Banned on government devices? | No | Yes, by several U.S. states and federal agencies |
Sources for each row are below.
Why the DeepSeek app is the problem
Your data is stored in China
DeepSeek's privacy policy says it collects, processes and stores your personal data in the People's Republic of China. That covers what you type, files you upload and your chat history.
That matters because Chinese law can require companies to hand data to the authorities. You have no say in it and no way to know.
That's why Texas banned DeepSeek on state-issued devices on January 31, 2025 (AP). New York, Virginia and several federal agencies, among them the Navy and the Commerce Department, followed with their own device bans.
It has had real security failures
- An open database. In January 2025, Wiz Research found a DeepSeek database open to the internet with no password. It held more than a million log lines, including chat history and secret keys. DeepSeek locked it down after Wiz reported it (Wiz Research).
- Weak guardrails. Cisco tested DeepSeek R1 against 50 harmful prompts from the HarmBench benchmark and it blocked none of them (Cisco).
- Easy to hijack. In September 2025, NIST's Center for AI Standards and Innovation (CAISI) tested DeepSeek's models against U.S. models. It found AI agents built on DeepSeek far more likely than agents built on U.S. models to be hijacked into following an attacker's instructions, and the models lagged on most performance tests (NIST CAISI report).
None of these alone means "never touch it." Together, they mean the app isn't a place for client data.
What about "DeepSeek is open source"?
True, and it's the one real point in its favor.
DeepSeek publishes its model weights, so anyone can run the model on their own hardware or in someone else's cloud. Amazon offers DeepSeek-R1 as a fully managed model in Amazon Bedrock, and Microsoft added it to Azure AI Foundry. Run that way, your prompts go to Amazon or Microsoft under their terms, not to DeepSeek's servers in China.
But:
- That's a developer setup, not an app your office manager opens.
- The guardrail and hijacking problems come with the model, wherever it runs.
- Per-use billing on those platforms can creep up if nobody watches it.
For most 1-20 person businesses, it's more work than it's worth.
What ChatGPT gets right (and where to be careful)
On ChatGPT Business, Enterprise and Edu, OpenAI says it doesn't train its models on your workspace data by default (OpenAI: Enterprise privacy). You also get an admin console to add and remove users, and Enterprise and Edu (with Business rolling out) can choose where workspace content is stored.
On the Free, Go, Plus and Pro plans, your chats can be used for training unless you switch it off:
Settings > Data controls > Improve the model for everyone (turn it off). OpenAI's Data controls in ChatGPT article explains what it covers.
Two edge cases:
- Turning training off isn't the same as not storing. Your chats are still kept on OpenAI's servers until you delete them.
- A personal account on a work task is still a personal account. If staff use their own ChatGPT login for client work, you have no admin control and can't remove their access when they leave.
And if you already pay for Microsoft 365 or Google Workspace, check what's included before buying anything. Microsoft 365 Copilot Chat with a work account shows a green shield when enterprise data protection is on.
Which is "better" at the actual work?
It depends on the task, and the answer changes every few months as new versions come out. Tom's Guide's early-2025 test had DeepSeek writing the livelier short story and ChatGPT explaining code more clearly. CAISI's September 2025 tests had the best U.S. model ahead on almost every benchmark.
For a small business, quality is rarely the deciding factor anymore. All the big tools write a decent email. The deciding factor is what happens to the information you give them.
Keep it safe in your business
- Approved list. Name the AI tools staff may use for work, and on which account.
- One-line rule for everyone: "No client names, account numbers, health or financial details in any AI tool that isn't on the approved list."
- Block what you don't allow. If you manage your devices or DNS filtering, you can block the DeepSeek app and website on company devices.
- Use company accounts, not personal ones, so you can remove access when someone leaves.
When to call someone
- Someone already pasted client or employee data into DeepSeek or another unapproved tool.
- You handle health, legal or tax information and need to know which AI tools your obligations allow.
- You want DeepSeek or another open model running privately and need it set up and watched.
If any of those sound like you, we can help.
Related reading
- Which AI tools can employees safely use at work?
- Are free AI tools safe for client data?
- AI vendor checklist: 12 questions before you share data
Originally published February 2025. Updated October 2026.





No comments yet