Is it safe to put your data into AI?
When someone asks me whether it is safe to put their information into AI, I start with a simple rule.
Never enter confidential, client, personal or private business information into Non-Private AI.
I call this the Golden Rule of AI Data Privacy.
But there's a question behind that rule. What makes one AI system private and another non-private?
To understand that, it helps to look at three common ways AI is provided: consumer services, business services and locally hosted systems.
They offer different levels of control, but none is automatically approved for every kind of information.
Three common AI environments
These are not ranked from least to most safe. A poorly secured local model is not automatically safer than a properly managed business service. And changing a privacy setting in your personal account does not automatically make it an approved work tool.
Consumer AI - the public square
Personal AI accounts and public-facing tools, such as the everyday versions of ChatGPT, Claude or Gemini. They can be useful, but shouldn't be assumed suitable for confidential business information.
Normally Non-Private AI for work, unless specifically approved.
Business AI - the rented office
AI provided under business arrangements with privacy, security and data-handling protections. Organisations must check those protections and approve the intended use.
Can qualify as Private AI.
Local AI - the vault
AI running on computers or servers under your control. This can avoid sending information to an external model provider, provided the rest of the system also keeps it protected.
Can qualify as Private AI when properly secured and approved.
Running a model locally gives you control over the model. Control over the whole system also means the logs, storage, backups and other tools it connects to. Where Does It Run? covers that in more detail.
Which AI should you use for confidential information?
Start with the Golden Rule. Never put confidential information into Non-Private AI.
If your organisation has approved a Private AI system, you can use it for the types of information it permits.
For public information or completely fictional examples, you have more freedom.
The decision isn't simply whether the AI runs locally or in the cloud. It's whether the system is properly protected and approved for the work you're doing.
What your organisation needs to check
Deciding whether a system counts as Private AI is the organisation's job, not every employee's. The checks include whether your information is used for training, how long it is stored and who can access it, how personal data is handled under GDPR, whether any transfer outside the EEA rests on valid arrangements, and what other services the tool connects to.
A signed contract is an important part of that, but it isn't the whole answer. The Vendor Evaluation Card helps you check what a tool really offers.
Be careful with the word "anonymous" when you test a tool. Changing a customer's name to "Customer A" doesn't necessarily make the information anonymous, because other details might still identify them. Completely made-up information is a much safer starting point.
Summary
Consumer, business and local AI can handle your information in different ways.
But you don't need to become an expert in those differences to use AI responsibly at work.
Your organisation should check which systems offer suitable protections and approve them for specific uses. Your responsibility is to use those approved systems and follow the rules.
That's the Golden Rule in practice: never enter confidential information into Non-Private AI.
This is an operating model, not legal advice. For a real deployment, confirm the GDPR specifics with a data protection adviser.
Put it to work: the Golden Rule of AI Privacy card is the one-page rule; the What Counts as Private AI? card sums up the three environments; and the Vendor Evaluation Card helps you check whether a tool is really Private AI or just says it is.