Someone in your business opened a browser tab this morning, pasted in a client contract, and asked an AI tool to summarise it. It took thirty seconds and probably saved them an hour. Nothing about it felt like a security decision to them, and in most businesses nobody else would know it had happened.
These tools are frictionless. They are simple, quick, and useful, and that ease tends to lower the care people usually bring to company information. Staff start uploading content without thinking much about it, and over time the habits they built around what should and should not leave the business begin to erode, inside AI tools and outside them. So if you wanted to see what your staff are putting into AI tools today, could you?
In a 2025 study by the University of Melbourne and KPMG, 44% of the Australian employees surveyed said they had used AI at work in ways that contravene policies and guidelines (Trust, attitudes and use of artificial intelligence: A global study 2025).
Nobody gave staff a safe way to do this
Most businesses have a policy that says be careful with AI. Far fewer have a way to tell whether it is being followed. Staff reach for whichever tool is open, and a lot of that use runs through personal accounts that sit outside the company's logins and controls. Blocking the tools outright tends to move the same behaviour onto personal devices, where you lose sight of it altogether.
In my experience shadow AI starts as a visibility problem. A written AI policy is hard to stand behind when you have no way to see whether anyone is following it, and the policy becomes useful once you can.
Where the prompt can be seen
If you stop and think about what happens inside a browser during a working day, it is most of the job: visiting websites, logging in with credentials, working in SaaS applications, uploading and downloading files, and now typing prompts into AI tools. Much of that AI use happens in a text box in a browser tab, which makes the browser a natural place for a control to see what goes into it.
A browser security layer works at that point. The more capable ones record the prompts staff send to AI tools, whether the login is personal or corporate, and whether the tool is a well known one or one of the lesser known models people are quietly trying out. Those prompts land in one place your leadership can search, which gives you a picture of AI use across the business that very few organisations have today.
Once that record exists, the same layer can apply your own data patterns, masking or blocking something like a card number or a client identifier before it leaves the browser. It has to do this without slowing anyone down. When a security control adds friction to the everyday CRM or website, staff push back and find a way around it, and the security team ends up on the back foot.
Most of what surfaces is well intentioned
When businesses first look, very little of what turns up is malicious. It is usually a capable person moving quickly, pasting something in to save time, with no sense that the material was sensitive or that it was leaving the business. There are smaller groups who know they should not and do it anyway, and occasionally someone who means harm. A browser layer gives you a view of all three, which makes the conversation about how staff should use AI a good deal easier, because it starts from what is happening.
See it in your own environment
You can't manage what you can't measure. The first step we suggest is a monitor-only pilot across ten to twenty users for a couple of weeks, with no policy changes and nothing blocked. It shows which AI tools are in use, through which accounts, and what kind of data is going into them.
In my experience that evidence changes the discussion faster than a survey or a policy review, because the data is your own. From there, deciding what to allow, mask, or block is a conversation with the facts in front of it.