Is My Data Safe?

Lesson 9 · taught by Pip

Here's the plain idea: "Is AI safe?" is the wrong question. The right one is "How is this particular assistant set up?" Safety isn't a property of AI in general — it's a property of the choices someone made when they built and connected it.

Think of an assistant like a new hire on their first day. The same person could be a careful, trustworthy employee or a liability — it depends entirely on what you hand them. Which files can they open? Which drawers stay locked? Are they allowed to wire money, or only to answer the phone? You decide that when you set up the role, not after something goes wrong.

So to judge whether your data is safe, you ask three things about the setup.

What does it store, and where does it go?

When you type into an assistant, that text travels to a company's computers to be answered. The honest questions are: does that text get kept, and is it ever used to train future versions?

Reputable business tools answer this plainly in writing. Many offer arrangements where your conversations are not used to train the model and are deleted after a short window. Consumer freebies often reserve the right to do more. The difference isn't the AI — it's the contract and the configuration behind it.

For a branded business assistant like the ones Portmint builds, the knowledge it draws on is set up deliberately. Public-facing facts — hours, services, FAQs — are fair game. Sensitive records aren't poured into the open part of the assistant; they're reached through a separate, locked-down connection only when truly needed.

What is it allowed to do?

Reading is one thing. Acting is another. An assistant that can only look up an answer carries far less risk than one wired to change a booking, issue a refund, or delete a record.

The safe default is simple: anything that moves money or erases data stays switched off until you explicitly turn it on, one capability at a time. A good setup also keeps a log of every action — so if you ever ask "who did that, and when?", there's an honest answer waiting. Least privilege, written down.

Who can see the conversation?

Last one: who's on the other end of the line? A widget on your public website is meant for customers and should be scoped to your business only. An internal assistant that sees private information should sit behind a login, visible only to your team. Mixing those two up is how data leaks — so the setup keeps them firmly apart.

Your turn

Pick one assistant you use or are considering. Ask the vendor two questions in writing: "Is my data used to train your model?" and "What actions can it take on my systems by default?" If the answers are vague, that vagueness is your answer.

You won't become a security expert from one lesson — but you now know the three setup questions that separate a safe assistant from a risky one: what it stores, what it's allowed to do, and who can see it.

Next, we'll turn that same clear eye on the marketing — judging AI claims without the hype. 🔦