Staying safe: security and your data
This sounds like the scary chapter. It is not. It is a short list of sensible habits, and once you have them, you can relax and get on with building.
People get nervous about putting anything near an AI tool, and that is fair. You should know where your information goes. There are only a few rules and they are simple. Learn them once and the worry mostly disappears.
What actually happens when you type
Claude is always connected to the internet. When you type a message and hit send, that message is sent off to Anthropic's computers, which work out an answer and send it back to you. That is just how it works. There is no offline version sitting on your laptop doing the thinking.
So here is the plain version: anything you type into the chat leaves your computer. Treat it that way. Not because anyone is out to read your messages, but because the safe habit is to assume that whatever you send, you have sent.
Treat anything you type to the AI as leaving your computer. The conversation goes off to Anthropic to get your answer back. If you would not be comfortable with that information going out, do not type it in.
The bit that surprises people
Here is the crucial difference, and it is genuinely reassuring once it clicks. The conversation goes to the AI. The things you build do not.
When Claude helps you make a small webpage or a little tool, that thing runs on your own computer. The files sit in a folder on your machine. The data it works with stays on your machine too. That is what "building locally" means: local is your own computer, as opposed to somewhere out on the internet.
So a tool you build to sort your own spreadsheet does its work right there, on your laptop, without sending that spreadsheet anywhere. Your files and your data stay put unless you deliberately choose to send them somewhere. That choice is always yours, and you have to make it on purpose. Nothing sneaks out the back.
The conversation
What you type to Claude goes off to Anthropic to get your answer. So keep sensitive things out of the chat itself.
The thing you build
The webpage or tool runs on your own computer. Your files and data stay local unless you deliberately send them out.
Keep sensitive data out of the chat while you learn
While you are still getting the hang of this, keep your most sensitive information out of the conversation. No full health records. No years of bank transactions. Not yet. Start with data that is boring, not sensitive, or completely made up. You can learn everything you need to learn on dull data, and if it goes sideways, nothing is lost.
This is the same advice from Before you start, and it is worth repeating because it is the habit that protects you most. Boring data first. Real data later, once you understand exactly what the tool is doing.
Work and employer data
This one is easy to forget and important to get right. Do not paste confidential work information into an AI tool without checking your workplace's rules first. Client lists, internal documents, anything covered by a confidentiality agreement: check before you share. Many employers have a clear policy on this, and some have approved tools you are meant to use instead. A two-minute check saves a very awkward conversation later.
A few simple habits
Start with boring data. Made-up, dull, or non-sensitive while you are learning. Real and personal only once you trust what the tool does.
Never share passwords or keys. Do not type a password, a bank login, or a secret key into the chat. A "key" is a long password-like code that some tools use to connect to other services. Keep them out.
When in doubt, ask. If you are not sure whether something is safe to share, ask Claude "is it safe to share this?" before you paste it. It will give you a sensible answer, and asking first costs you nothing.
Check work rules before pasting anything from your job. If there is any doubt, treat it as confidential until you have confirmed otherwise.
So, is this safe?
Yes. Building this way is safe when you follow a few simple rules, and you now know them. This is about being sensible, not scared. You are not doing anything risky by making a little tool on your own computer. You are just being a tidy, careful person who knows where their information is going. That is all security really is at this level: good habits, calmly applied.
The short version
- 1Claude is always online. Anything you type is sent to Anthropic to get the answer back, so treat it as leaving your computer.
- 2The conversation goes to the AI. The things you build run on your own computer and stay there unless you choose to send them out.
- 3While learning, keep sensitive data out of the chat. Start with boring or made-up data.
- 4Check your workplace rules before pasting any confidential work information.
- 5Never share passwords or keys. If unsure, ask Claude "is it safe to share this?" first.
Sensible, not scared. Keep those habits and you can build with a clear head.