Safety · AI & Privacy
AI tools are extraordinary at handling information — which is exactly why it matters what you hand them. This page shows you what actually happens to your words, what should never go in the box, and how to keep working fast without giving away something you can't take back.
The short version
You don't need to understand the engineering. You need to understand these three, because everything else on this page follows from them.
When you press enter, your words travel to a server somewhere else and are processed there. Treat the prompt box like an email to an outside company — not like a private notepad.
Most providers store conversations so you can return to them, and keep copies for a period afterwards for safety and abuse checks. How long depends on the product, your plan, and your settings.
Free and personal tiers sometimes learn from what you type unless you turn that off. Business and enterprise plans usually don't by default. This is the single setting most worth checking.
Where it actually goes
Say you paste a paragraph from a contract and ask for a plain-English summary. Here is every place that paragraph exists afterwards — and where you have a say.
Test yourself
Eight things that land on a real desk. For each one, decide what you would do before it goes anywhere near a prompt box.
Know it when you see it
Almost everything people get wrong falls into one of these four buckets. Learn the shapes and you stop having to think about it case by case.
Information that identifies a living person who never agreed to this.
Things your employer or client would not want a competitor reading.
Categories where the law sets the rules, not your judgement.
The one bucket with no grey area at all. It simply never goes in.
Try it on real text
Paste anything you were about to hand an AI tool. This will point out the parts that look sensitive and offer you a cleaned-up version to use instead.
This runs entirely inside your browser. Nothing is uploaded, logged, or sent to this site — you can disconnect from the internet and it still works.
Settings worth checking
Menus move and wording changes, so learn the four controls rather than the click paths. Open your tool's settings and look for these.
Make it automatic
Tick them off as you build them. Once these are reflex, you stop needing to think about privacy every time you open a prompt box.
If it already happened
It happens to careful people. What matters is the next twenty minutes, not the mistake.
If it was a password, key or token, rotate it now. Everything else on this list can wait until that is done — a live credential is the only part that gets worse by the minute.
Remove the chat, and the uploaded file if there was one. It will not erase every backend copy, but it shortens the window and stops it resurfacing in your own history.
If it involved personal data or client material, your privacy or security contact needs to know. Regulated breaches often carry reporting deadlines measured in hours, and they start when it happened.
Then change what made it easy — the wrong account, training left on, a file that should never have been downloadable. Blame teaches nobody anything.
You now have the instinct. If you're rolling AI out to a team, the governance kit turns these habits into something you can actually hand people.
