Who owns the data your AI uses?

There is a category of product that asks you to hand over the most valuable thing you have, your accumulated professional knowledge, in exchange for convenience. Most people do it without reading the terms. That was a survivable mistake when the thing being uploaded was a photo. It is a different mistake when it is twenty years of client work.

Three questions settle it. They are worth asking of any AI product, including ours.

One: does my material train your model?

This is the question that matters most and is answered least clearly. There are three real answers and a lot of language designed to blur between them.

• No, never. Your material is used to answer your questions and nothing else.
• Yes, unless you opt out. Common, and the opt out is usually in a settings page you have to go find.
• Yes. Sometimes stated plainly, more often phrased as improving our services.

If your knowledge is your livelihood, only the first answer is acceptable. A model trained on your method can reproduce your method for people who never paid you for it.

Two: can I take it out?

Deletion has two meanings and companies rely on the gap between them. Removing a file from a dashboard is not the same as removing its influence from a system that has already indexed it, embedded it, or used it in training.

Ask specifically: if I delete a document today, is it gone from the index, and was it ever used in any training run? If the second answer is yes, deletion cannot undo it. That is a one-way door and it is worth knowing you are walking through it.

Three: where does the processing happen?

Location is not a technicality. It determines which laws apply to your material and who can compel access to it.

The honest picture is more complicated than most marketing admits. A product can host your files in one region and still send a query to a model that only runs somewhere else. Regional preference is real for some models and unavailable for others, with a fallback to US processing when no regional path exists. Any vendor telling you the answer is uniformly simple is either not thinking about it or not telling you.

What good answers look like

A vendor treating this seriously will tell you, without you having to dig:

• Your content is not used to train shared models, stated as a commitment rather than a default setting.
• What deletion removes and how quickly.
• Which processing happens where, including where it falls back.
• Who can see what, and how access is granted and revoked.

Why this is the whole argument for personal AI

The reason people held their own data was never abstract. It was that value accrues to whoever holds the asset. For twenty years the asset was behavioural data and the holders were platforms. With AI the asset is knowledge, and it is a much more concentrated kind of value: not what you clicked, but what you know.

If that knowledge sits inside someone else's model, the leverage sits there too. If it sits in a system you control, and only answers when you point it at someone, the leverage stays with you.

That is the entire design principle behind an AI Twin. Not privacy as compliance, but ownership as the thing that makes the product worth having.