Private AI: how to use AI without giving away your data
You can use AI with sensitive data — but a free chat app on its default settings is rarely the right place for it. There are five levels of privacy, from a settings change to models that never leave your own computer. Here is what each one protects, what it costs you in convenience, and how to choose.
Five levels of privacy
| Consumer app, settings changed | Business plan | API, your own account | Local model on your computer | AI on your own servers | |
|---|---|---|---|---|---|
| Who it suits | One person, non-sensitive work | Teams using ChatGPT or Claude every day | Automations and custom tools | One person with sensitive files | Companies whose data must stay inside |
| Used to train models? | Depends on your settings | Not by default, under business terms | Not by default, under API terms | No: it never leaves your computer | No: it never leaves your servers |
| Quality of the AI | The best models | The best models | The best models | Good, a step behind the best | Good, depends on your hardware |
| Effort to set up | Lowest | Low | Medium | Medium | Highest |
Based on the providers' published terms as of October 2026. Terms change, so read the current ones for the plan you use.
What never to paste into a consumer chat app
- Passwords, access keys and recovery codes.
- Customers' personal data.
- Contracts, financials or anything under a confidentiality agreement.
- Health information — yours or anyone else's.
- Anything you would not send by email to a stranger.
AI on your own computer
Free tools such as Ollama and LM Studio run open AI models on a laptop or desktop. A recent computer handles the smaller models well. They are slower and a step behind the best cloud models, but nothing you type or open leaves the machine. That makes them a good fit for summarising private documents, drafting from your notes, and anything you would hesitate to put in the cloud.
For a company
Most companies need three things: business plans of the AI tools people already use, so company data stays out of personal accounts and out of training; clear rules on which data may go into which tool; and, for the most sensitive data, models running on the company's own servers. Rules only work if people know them, so they come with a short training.
When it is worth getting help
- Your team already uses AI with company data, and nobody has set rules.
- You want a local or on-premise setup and do not know what hardware or model to choose.
- You want an assistant that works with your own files without sending them to the cloud.
Frequently asked questions
Is ChatGPT safe for company data?
The business plans of ChatGPT and Claude do not use your data to train models by default, under their published terms. The free and personal plans depend on your settings. For company data, use a business plan and agree which data may go into it.
Do ChatGPT and Claude train on my chats?
On personal plans it depends on your privacy settings, so check them. On business plans and through the API, not by default under the providers' current terms.
Can I run ChatGPT on my own computer?
Not ChatGPT itself, but open models that do much of the same work run locally with tools like Ollama or LM Studio, and nothing leaves your machine.
Is local AI as good as the cloud?
Not quite: the best models run in the cloud. For summarising, drafting and working with your own documents, local models are often good enough, and the privacy is complete.