Private AI for business: ask who else is on the server, not where it is

When a managing director asks about AI and data privacy, the first question is almost always the same: where are the servers? Which country, which region, which set of laws? It is a reasonable question. It is also, more and more, the wrong one.

The question that matters is simpler and less comfortable: who else is on the same system as us?

Most AI tools sold to businesses are shared by design. Your emails, your plans, your client history and your internal conversations pass through the same platform as thousands of other organisations. The vendor will point to encryption and access controls, and those controls are real. But they answer a different question from the one that decides your risk.

Why agentic AI raises the stakes

Ordinary business software stores data. Agentic AI, meaning AI that plans and acts rather than only answering, does much more than that. To be useful it has to read your email, sit in on your meetings, understand your documents and remember what was agreed. It builds a working picture of how your company thinks, who matters to it and what it has promised.

That picture is one of the most sensitive things a company will ever hold. It is also the thing that makes the AI valuable. The more the AI knows, the more it can do for you, and the more it matters where that knowledge lives and who can reach it.

This is why private AI for business is not a niche concern for regulated industries alone. Any company that wants its AI to do real work has to give it real access. The architecture behind that access decides how comfortable you can be.

The trouble with sharing

Shared platforms are efficient, and for many kinds of software they are perfectly sensible. For an AI that holds your company memory, sharing creates risks that encryption alone does not remove.

• A wider blast radius. On a shared platform, a weakness found in one place can matter to every customer at once. You are exposed to problems that start in someone else's account.
• Blurred edges. When many companies' information is processed side by side, the line between one company's knowledge and another's depends entirely on software rules being right every time.
• Memory you do not hold. Over time your AI builds a detailed picture of your business. On a shared platform, that picture sits in a system arranged around the vendor's needs rather than yours.
• Harder questions to answer. When a client, an auditor or a board asks exactly where your information goes and who could see it, a shared platform makes the honest answer long and conditional.

Geography is a separate question

None of this means location is irrelevant. Data residency rules are real, and some companies need their information held in a particular place. Ask where your system will sit and why.

But geography and separation are different questions. A system in your own country, shared with a thousand other businesses, does not give you control. A private server dedicated to your company, kept apart from everyone else, gives you far more, wherever it happens to sit. Whatever the location, insist on separation.

What private AI should actually mean

The phrase private AI is used loosely, so it is worth being precise. For a company, it should mean that the AI runs on a private server dedicated to that one company, with its own accounts and its own AI account, walled off from every other customer. Nothing about another company's work should ever be mixed with yours.

Privacy also has an inside. Inside a single company, not everyone should see everything. A good private AI holds one memory for the whole business, but shows each person only what their role allows. The finance team's numbers, the board's discussions and a manager's notes on their team should reach the right people and no one else, even when someone asks the AI directly.

And privacy is not only about reading. An AI that acts can also send, book and spend. Private AI for business should hold anything that goes out or costs money until a person approves it, and should keep a record of every action and the reason for it.

Questions worth asking any provider

If you are weighing up AI for your company, these questions will tell you more than any brochure.

• Does our AI run on a private server dedicated to our company, or on a platform shared with other customers?
• Does it have its own accounts and its own AI account, separate from every other company?
• Inside our company, how do you make sure each person sees only what their role allows, and how was that tested?
• What can the AI do on its own, what waits for a person, and can we see a record of every action?
• Is our company walled off from every other client, and are we tied to a single AI provider?
• If we leave, what happens to our company's memory?

If a provider cannot answer these clearly and specifically, that is an answer in itself.

How Anabel keeps company AI private

Every company that uses Anabel runs on a private server dedicated to that company, which we provide and manage, with its own accounts and its own AI account, walled off from every other company. We install her, teach her the business and run her, and she is never tied to one AI provider. Inside the company, each person sees only what their role allows, anything that goes out or costs money waits for approval, and every action is logged with its reason.

We test that wall rather than assume it. On a full company setup we made more than 160 deliberate attempts to reach information a person should not see, and none succeeded. We run our own companies on Anabel, so the privacy we promise is the privacy we rely on ourselves.