Your Data Stays Yours: How Our AI Assistant Keeps Your Business Information Safe

Your Data Stays Yours: How Our AI Assistant Keeps Your Business Information Safe

When we talk to companies about using AI with their own data, we hear the same worry almost every time: "We would love this. But we do not want our data going to an outside AI company."

That is a fair concern. Your data is your money, your customers, your orders and your people. So here are the six questions we are asked most often, with plain answers. 

1. "Will the AI learn from our data and share it with others?"

No. We use the business versions of AI services, not the free chat apps anyone can open in a browser. The companies behind them, including Anthropic and OpenAI, say that they do not use data from business customers to train their AI by default.

For companies that want the strictest setup, the AI can also run inside your own company's private space in the cloud, for example through Amazon Bedrock. Amazon says that in this setup your questions and answers are not shared with the AI company and are not used for training. Some providers also offer a "keep nothing" option, where they do not store your questions at all. It is available to eligible large customers.

The short version: you choose the level of protection that fits your rules, and we build it that way.

2. "Does the AI get to see our whole database?"

No. This is the most important point in this post.

Think of the AI as a translator. It does not hold your data. It only helps your own system understand a question. Here is what happens:

  1. Someone asks a question in everyday words.

  2. The AI turns it into a request that your own database understands.

  3. Your own database does the work and finds the answer, on your own systems.

  4. Only the small answer needed comes back, and the AI writes a short summary of it.

The AI never takes a copy of your data. The facts stay where they always were.

For extra protection, personal details such as names, emails and phone numbers can be hidden before anything reaches the AI.

3. "Can people see things they should not?"

No. The assistant follows the same permissions you already have. A team member sees only their own area. A regional manager sees only their region. The head office sees everything. Whatever your rules are today, the assistant keeps to them.

What matters is where this is checked. The check happens behind the scenes, for every single question, and not only in the chat window. It works like a locked door. Asking in a clever way does not open it.

4. "Can the AI change or delete our data?"

No. The assistant can only look and report back. Think of it as a window, not a door. It cannot edit a record, remove a customer or touch an order or an invoice.

Later, the assistant may be able to take actions, such as creating a follow-up task. If that happens, it will come with approval steps, a record of everything that was done and a way to undo it. Never as a side effect.

5. "Will it make up numbers?"

This is the worry people forget. A confident wrong number is worse than no number at all, whether it is a revenue total, a stock count or the number of employees.

Our rule is simple: your database does the maths, and the AI only explains the result in words. Because the numbers come from your own data, the same question gives the same answer every time.

To keep it that way, we keep a fixed list of test questions and run them again whenever the AI or the data changes. If an answer drifts, we notice before you do.

Predictions, such as forecasts and risk scores, work the same way. They come from proper models that are tested for accuracy, and not from asking a chatbot to guess. The AI explains the prediction. It does not invent it.

6. "Who looks after the passwords and keys?"

This part is less exciting, but it matters. Our standard way of working is:

  • passwords and access keys are kept in a secure place, never written into the software itself

  • sensitive details are hidden in our records of what the system does

  • every connection to an outside service is recorded, so it can be checked later

It is basic care, and we do not skip it just because AI is the exciting part.

Security comes first, not last

On our own project, we built the safety foundations before the AI was even switched on. That meant the rules about who can see what, the safety checks and the tests. It is like putting locks on the doors before you move into the house. Adding them later is much harder, and things tend to slip through.

Have a security question we did not answer here? Ask us. We would rather answer it in the first meeting than have it worry you for months. And next week, we will show how a project like this actually starts.

Missed Part 2? Read Talk to Your Data: Ask, Understand, Predict