Nobody Types Personal Data on Purpose

When I paste a long email into an AI chat, what I want back is a summary or a cleaner draft. Handing over someone's personal details is not the point of the exercise.

The details are part of the text

What you paste is usually a piece of work in progress. A thread with a client. An enquiry that arrived through a form. A log that threw an error. A clause out of a contract.

None of that registers as "a file containing personal data". It registers as text you want shortened, or text you want fixed. But the name at the top comes along with it, and so does the signature, the phone number, and someone else's address sitting inside a quoted reply.

Very few people set out to type personal data. They type a piece of work, and the data was already in it. The order matters more than it sounds.

Pasting is also typing

While it sits on the clipboard it is still yours. What changes is the moment it goes to a service outside your machine.

"Paste" sounds like sliding a sheet of paper across a desk. What actually happens is that text under your control gets entered somewhere that is not. The same action, carried by a lighter word.

What gets missed sits at the edges

In a long piece of text, attention goes to the body, because the body is the part you want summarised.

What I miss is at the edges. The signature. The older exchange quoted underneath. The corner of a table that came along with the copy. An identifier at the tail end of a log. While you are reading the body, none of that is in view.

I do not think that is a matter of how careful someone is. It is a matter of where the care is pointed. In my case, the thing I notice afterwards is always at an edge.

Training is not in the room at the moment you send

Explain how to handle information once and it lands as knowledge. But the pasting happens on an afternoon with a deadline close, or while you are hurrying a reply. Whether any of it comes back to you at that moment turned out to be a separate question from how much you know.

There is a state where you know and still miss it. If that is what is happening, the thing to add is not more explanation. It is something visible at the moment of sending. That is what Safe Privacy Gate is for: just before the text in an input field goes out, it puts the likely personal details on screen. That is the whole of it.

Detection is sometimes wrong, and there is no single right answer

Two reasons it stops short of deciding anything for you.

One: detection gets things wrong. A run of digits can look like a phone number, and a name can fail to look like a name. Mistakes happen, and a tool that blocks automatically stops your work every time it misreads something.

Two: whether a thing is fine to send depends on the purpose of the text and on who is receiving it. Typing your own name is usually not a problem. The same shape of data, belonging to someone else, is worth pausing over. An internal tool and an outside service are not the same either. A tool cannot see where that line falls.

So it does not block, and it does not quietly delete. It shows what it found and leaves the choice with you, right there — send as it is, or fix it first.

I did not build it because I distrust the people using it. I built it because even someone being careful will struggle to read every part of a working document, every single time.

The hard part was never the willingness to check. It was having somewhere to do the checking. So that is the only thing I added.