It's leaking from the tab your team already has open.
Everybody says the same two things about AI and your data. Be careful. Use guardrails.
Nobody says what either of those means.
So you are left holding a warning with no instructions attached, which leaves most business owners doing one of two things. Ignoring it completely and hoping for the best. Or freezing, deciding AI is too risky to touch, and waiting until somebody explains it properly.
I want to explain it properly, because the fear is pointed at the wrong thing.
The part everybody gets backwards
Here is the sentence that sounds alarming and is completely normal.
Your data goes into AI.
Of course it does. That's the entire point of it. AI cannot help you with your business if it knows nothing about your business. Feed it nothing and you get generic, confident, useless answers, which is exactly why so many people try AI once and conclude it's overhyped.
So the question was never whether your information goes in. The question is whose AI it goes into.
That distinction is everything, and almost nobody makes it.
When we set AI up for a business, their information goes into it. Deliberately, into their own account, in their own environment, with settings they control and a history they can go back and read. That isn't a leak. That's a business using its own information on purpose.
When somebody on your team pastes a client contract into a free tool on their phone at nine at night, the same information goes somewhere nobody owns, on an account nobody manages, with retention settings nobody has ever read. It's the same data and it's a completely different exposure.
One of those is a system. The other is a risk somebody took on your behalf, without telling you.
## What we actually do here
We had to answer all of this for ourselves, because we run our own company on it. Here is where we landed.
Our AI has no standing access to anything we don't deliberately give it access to. It cannot reach into our systems and help itself.
What we hand it is anonymized by default. Not because a policy told us to, but because the identifying detail is almost never necessary. The AI does not need to know which client it is to help me think through the work.
Now, don't get me wrong, my AI does have access to my calendar, email and Notion database but does not have rights to send email or permanently delete any of my information. That's guardrails at work.
The training setting is off on every AI tool we use. Our information is not used to improve anybody's model.
That's not a sophisticated setup. It took an afternoon of decisions and it's the difference between using AI and wondering about it.
What a governance framework actually asks
You do need one. Those words sound like the title of a binder nobody reads, but underneath it's not complicated, and it starts with three questions you should be able to answer about any tool anyone at your company is using.
What can it see? What information actually goes in, and is any of it something you would not email to a stranger.
What can it reach? Does it have standing access to your files, your inbox, your systems, or does it only know what somebody deliberately gives it.
What can it do on its own? Anything that touches money, client data, or something you cannot undo should have a person in front of it. Not because AI is untrustworthy, but because that's how you would treat a new hire in week one.
Those three are the beginning, not the whole of it. But if you cannot answer them about the tools already in your business, including the ones you may not be aware of, no policy document is going to help you.
Where we started
When we worked through this for ourselves, two things came first.
The training setting. Most tools have a switch controlling whether your conversations are used to train the model, and in our experience almost nobody has looked at it. It's the simplest thing on this list and it's worth knowing where yours is set.
Whether we were using the right tool for the job. Deep research, design work, and storing your business knowledge so you can get real recommendations back are three different jobs. Reaching for the same tab for all three is how people end up disappointed and exposed at the same time.
Neither of those is the whole answer. They're just the first places we looked, and they're easy to look at.
The thing to actually stop doing
Stop assuming nobody on your team is already using AI.
They are. On their phones, on personal accounts, with your work in the prompt. Not because they are careless, but because it makes their day easier and nobody told them not to. That's the ordinary state of almost every business right now, including ones with a policy.
Which turns the whole fear around. Using AI is not the risk. Not deciding is the risk, because it's happening either way, just without you in it. Waiting does not keep you safe. It keeps you uninformed.
**So here is the thing to go and do today.** Ask your team what AI tools they have used and what they put into them. No blame, no policy, just the question. Most owners have never asked it once.
You cannot govern something you're not aware of. And you will almost certainly find out that the leak you have been worried about has been happening quietly for months, in the one place you were not watching.