An AI tool can only see what you give it. That makes keeping your data safe a handful of decisions you make, not a feature you buy.
Most owners we meet are in one of two spots. Either they’ve stopped using AI for anything real because they’re not sure what’s safe. Or the team is pasting whatever’s handy into a free chat tool and hoping. Both come from the same gap. Nobody has explained in plain words what the tool sees, what it doesn’t, and who’s in charge of the line.
What the tool can see
There are only three ways anything gets into an AI tool. Someone types or pastes it. Someone uploads a file. Or someone connects the tool to a system, like an inbox or a shared drive, and it reads from there on its own.
The first two are one-off. What you paste into a chat is what that chat can see, and nothing else. The third is ongoing. A tool connected to your inbox can see every email in it, including the ones you’d never have pasted in. It keeps seeing them until you turn the connection off.
Then there’s what happens to it afterwards. On a free consumer account, most tools can use your conversations to train their models unless you switch that off. On a business plan, that’s usually off by default and written into the contract. The setting takes a minute to find. In our experience, most businesses have never looked.
What the tool can’t see
It can’t see your drive, your accounting software or your CRM unless someone connects them. It doesn’t go looking. Nothing gets in without a person or a connection putting it there, which is good news, because it means the line is yours to draw.
What it also can’t do is tell the difference between a public brochure and a client’s medical history. To the tool, text is text. It has no idea a name is confidential or that a spreadsheet is your full P&L. The judgement about what’s sensitive stays with the person at the keyboard, every time.
Who decides
You do, and it helps to have a rule the whole team can repeat. Ours is the front page test. If you wouldn’t want it on the front page of the paper, don’t paste it into a tool you haven’t set up for it. Marketing copy, public information, your own process notes, your website, all fine. Client names with their details, financial records, health information, passwords, anything under an NDA: those stay out. The exception is an account you’ve set up properly for them and decided is allowed.
Two Australian things sit behind that line. The Privacy Act covers most businesses turning over more than $3 million, and some smaller ones regardless, like health providers. If it covers you, you’re responsible for how personal information is handled, whichever tool touched it. A leak may need reporting under the notifiable data breaches scheme. The government’s voluntary AI guidance for business is shorter than people expect. Its practices are what a careful owner would do anyway. Someone accountable, a human able to step in, honesty about where AI is used, and checking that it works.
None of that is legal advice. It’s the shape of the responsibility, and it sits with the owner.
Five rules you can set today
- Turn off data sharing on every AI account the business uses. Move anything with real work in it to a business plan.
- One login per person, with two-factor on. A shared AI account with a shared password holds every conversation anyone’s ever had in it.
- Names and numbers stay out unless the tool is set up for them. Where you need the tool’s help on a client matter, take the identifying details out first.
- Anything that pays money, changes bank details or goes to a client gets a person’s eyes before it goes. New bank details get confirmed on a phone number you already have.
- Connected tools get the least access that does the job, and they start read-only. Drafting comes later. Sending comes last.
Write it down so it outlives the tool
Rules that live in someone’s head leave with them. We keep the setup in three plain documents: who you are, how you work, and what’s happening now. The data rules above sit in “how you work”, next to how you quote and how you talk to customers. Every AI tool the business uses reads those documents first, and every new staff member reads them on day one.
That folder is yours. It doesn’t live inside any vendor’s app. When a tool changes its terms or a better one turns up, the rules and the context come with you. Trust in an AI setup comes from knowing exactly what it’s been given and being able to check.
If you’d like that folder built with you, set up your own AI Operating System takes you through the three documents. If you’d rather we built it into your real systems, work with Brightwaters.