Journal · The AI Architect · 2026-07-19
Never paste passwords or account numbers into a prompt. If a task seems to need one, that's a signal to pause.
Never paste passwords or account numbers into a prompt. If a task seems to need one, that’s a signal to pause.
The tension isn’t really about security hygiene. Most people already know not to shout their bank PIN across a room. The tension is that AI tools are designed to feel like a private conversation, and that feeling makes us forget we’re often talking to a system that logs, stores, and sometimes learns from what we type. The real problem is a mismatch between how safe a prompt window feels and how safe it actually is.
The interface lies to you, a little
A chat box looks like a diary. It’s just you and the cursor, no audience, no witnesses. So when a task genuinely seems to require a password — “log into my account and check this for me,” “here’s my routing number, format this invoice” — it feels no different than jotting a note to yourself. But that box is not a notebook. Depending on the tool, your input may be stored, reviewed by humans for quality control, or used to train future models. Even when a company promises otherwise, you’re trusting a policy you didn’t write and can’t verify in the moment.
This isn’t a reason for paranoia. It’s a reason to treat the prompt window the way you’d treat a text message to a stranger you’ve never met, even if that stranger seems helpful and speaks in full sentences. You wouldn’t text your debit card number to a helpful stranger just because they offered to help you balance your budget.
Try this: before you paste anything sensitive into a prompt, ask out loud, “Would I read this over the phone to a call center employee I’ve never spoken with before?” If the answer is no, don’t paste it.
Treat “it needs my password” as a design flaw, not a requirement
Here’s the more useful reframe: if a task seems to require your actual credentials, that’s rarely a fact about the task. It’s usually a sign that you’ve reached for the wrong tool, or that the tool is asking you to do something it shouldn’t.
A well-built AI assistant doesn’t need your Netflix password to recommend what to watch next. It doesn’t need your bank login to help you build a budget spreadsheet. If you find yourself thinking “I’ll just give it access so it can check this directly,” pause and ask what the AI actually needs — usually it’s information about your situation, not the keys to your accounts. You can say “my account has three subscriptions totaling $47 a month” instead of logging in through the chat. You can copy a de-identified snippet of a document instead of forwarding the whole thing with your account number in the header.
This distinction matters because it shifts the responsibility. You’re not the one being careless for wanting help. The task, as framed, is asking for more trust than it has earned. That’s worth noticing, and it’s worth treating as a stop sign rather than a minor inconvenience to route around.
Try this: the next time you’re tempted to paste in a password “just this once,” rewrite the task instead. Describe the outcome you want and the relevant details, stripped of anything identifying or sensitive. Nine times out of ten, the AI can still help — it just needs a translated version of the problem, not the raw material.
Build a small habit, not a big rule
Rules people don’t follow are worse than no rules at all, because they create false confidence. So instead of memorizing a long list of “never do this,” build one small habit: keep a mental folder labeled “things that never go in a prompt.” Passwords. Account numbers. Social Security numbers. Full addresses tied to financial accounts. Anything with a security question answer in it.
When something in that folder shows up in a task, that’s your pause button. Not a full stop necessarily — sometimes the answer is simply to remove the sensitive part and continue. But it should always interrupt the flow enough for you to think, even for three seconds, about what you’re about to send and where it’s going.
The goal isn’t to make you anxious every time you open a chat window. It’s to make the pause automatic, the way checking your mirrors becomes automatic when driving. Once the habit is built, it stops feeling like friction and starts feeling like common sense.
This is a small piece of a much bigger conversation about trust, tools, and where the boundaries should sit between convenience and exposure. The book goes further into how to think about that boundary across different tasks and tools, but the habit itself — noticing the pause, and honoring it — is something you can start today, with the very next prompt you type.
Go deeper. The full method is in The AI Architect. New here? Start with the free companion pack, or explore the series.