Responsible AI: Privacy, Verification, Limitations & When NOT to Use AI

Introduction

You've built real, everyday confidence. You know what AI is and isn't, how it builds an answer, why it slips, the shape of the AI landscape, and how to team up with it. This module is the one that keeps all of it safe: three calm habits — protect what's private, check what matters, and know when a decision belongs to a qualified person, not a tool. None of this comes from fear. It's the quiet confidence of someone who uses AI well and stays firmly in charge. By the end of these 30 minutes you'll be able to say, about almost any task: I know what to keep private, I know what to check, and I know when to let a qualified person decide.

What to keep private

After this lesson you will be able to decide what is safe to share with an AI tool, and what should stay private.

A short story. Mei keeps the books for a few small shops. One busy afternoon she was about to paste a client's full financial records — names, account numbers, the lot — into an AI tool to "tidy them up faster." She paused. She didn't actually need to reveal who the client was to get the formatting help she wanted. So she removed the private details, kept only the plain structure, and got exactly the help she needed — without handing over anything sensitive.

Here's a simple rule. Treat anything you type into an AI tool with the same care you would use before sharing it digitally. If you're unsure, assume it should be protected. So keep private things private. One kind of information is never worth sharing: passwords, PINs, and security codes — a genuine AI task never needs them, so don't enter them into a tool at all. And here is a short list of other things to keep out unless you have a clear, authorized, and necessary reason to include them:

• •Bank, card, and account numbers.
• •Medical details — about yourself or anyone else.
• •Another person's personal information — a customer, a patient, a child, a colleague.
• •Confidential work information your organization hasn't made public.
• •ID numbers, home addresses, and other details that identify a specific person.

Share the task, not the secrets

Here's why it's easy: AI is a wonderful helper for the shape of a task — drafting, organizing, summarizing — and you rarely need to reveal sensitive details to get that help. For many low-stakes drafting, organizing, and summarizing tasks, you can remove or replace private details and still get useful help. Redaction reduces exposure; it does not guarantee anonymity or make sensitive content safe to share.

A helpful analogy. It's like talking through a problem in a busy café. You'd happily ask a friend how to word a difficult letter — but you wouldn't read your bank details out to the whole room. Same idea with AI: share the task, not the secrets.

Your turn — which is safe to paste into an AI tool? (a) A customer's full name, address, and card number. (b) "Help me write a polite payment-reminder message," with no personal details. (c) Your email password, so it can "log in for you." Answer: (b). Share the task and the wording you need; keep identities, numbers, and passwords out.

Here's the aha: keeping information private isn't about distrust — it's simple, everyday care, and it costs you almost nothing.

A quiet reminder: being careful doesn't make a tool bad or unsafe — it keeps you in control of what leaves your hands. And if you're ever unsure whether something is private, treat it as private.

A practical example. Mei needs a firm-but-kind overdue-invoice message. She writes "a client" instead of the name, leaves out the amount and account number, and asks only for the tone. Two minutes, genuinely useful, nothing sensitive shared.

A common mistake to avoid: pasting a whole real document "just to save time" without removing the private parts first. Take ten seconds to redact — remove or replace names, numbers, and identifying details — before you ask.

What you can do today: think of one thing you'd never say out loud in a crowded room; that's a good sign it doesn't belong in an AI prompt either.

What you can do now: name three kinds of information you'll always keep out of an AI tool.

Trust, then check

After this lesson you will be able to check an AI answer before you rely on it — and tell when a confident answer still isn't enough.

A short story. Idris volunteers at a small community radio station. An AI tool gave him a neat, confident-sounding detail about a local road closure, and he almost read it on air. Something felt a little too tidy. He made one quick call to the council to confirm — and the detail was out of date. One check kept the wrong information off the air for the whole neighborhood.

Let's bring back two friends from earlier modules. The AI Confidence Trap™ (Module 3): AI can sound completely sure and still be wrong — confidence is not proof. And the Everyday Verification Core (Module 3): for anything that matters, check what matters — is it true, is it current, does it fit my situation? A tool produces an answer from learned patterns; when the answer really matters, you ground it by checking against a trusted source. And remember from Modules 4–5: a shown source is evidence to inspect, not proof on its own.

So here's the everyday shape of it:

• •When AI is genuinely helpful: drafting, brainstorming, organizing, summarizing, explaining something in plain language, giving you a starting point.
• •When AI is not enough on its own: anything you'll act on where being wrong would cost time, money, safety, or someone's trust — facts, figures, names, dates, instructions, and advice you'll rely on.

Limits you can't always see

A helpful analogy. It's like a very confident acquaintance who's often right but sometimes mistaken. You'd happily hear them out — then check anything important before you act on it.

Your turn. An AI tool gives you a confident answer you're about to rely on for something important. What's the responsible next step? (a) Use it right away — it sounded certain. (b) Decide it must be wrong and ignore it. (c) Check what matters against a trusted source before you act. Answer: (c). Confidence isn't proof; a quick check keeps you in charge.

Here's the aha: you don't have to check everything — just the parts you'll actually rely on. Small habit, big protection.

A word on limitations: AI also has limits you can't always see — it can be out of date, miss your local context, or reflect patterns that aren't fair or complete. It can also misunderstand your situation or confidently fill a gap with something that simply sounds reasonable. Checking doesn't make an answer perfect; it keeps you the one deciding whether to trust it.

A practical example. Idris uses AI to draft the show's running order (a great use), then confirms the one time-sensitive fact with the council before reading it out (a necessary check). Draft from AI; verification and decision from Idris.

A common mistake to avoid: letting a polished, confident answer skip the check — "it sounded so sure." Smooth wording is not evidence. Keep the check for anything that matters.

What you can do today: next time an AI answer feels a bit too tidy, do one quick check before you use it.

What you can do now: say the two-part habit in your own words — trust the help, then check what matters.

When to let a person decide

After this lesson you will be able to recognize the situations where a decision should always involve a qualified person — and use AI only to prepare, never to decide.

A short story. Rosa cares for her elderly father. When he had a worrying symptom, she used an AI tool to gather her thoughts and write down clear questions to ask — then she took him straight to a doctor. AI helped her feel prepared and calm; the diagnosis and the decision came from a qualified professional, exactly where they belonged.

Some decisions carry real weight — for health, safety, money, the law, or a vulnerable person — and those should always involve qualified people and official channels. AI can help you prepare — gather your thoughts, list questions, explain a word in plain language, organize what you already know — but it must not make the call. Keep this short list of "a person must decide" moments:

• •Emergencies and dangerous situations — contact the proper emergency services, now.
• •Medical concerns and diagnosis — see a qualified health professional.
• •Legal matters and important rights — talk to a qualified legal person or official service.
• •Money and investment decisions — use a qualified, regulated professional.
• •The safety or welfare of a child or a vulnerable person — go to the proper people and official channels.
• •Any decision that is hard to undo — slow down and involve the right person.

Why these stay with people

Here's why these stay with people: they need real accountability, professional training, up-to-date judgment, and a duty of care that a tool simply cannot carry. (Remember Module 5: capability is not authority. And Module 3: responsibility can't be uploaded.) AI can inform; a qualified human decides and is answerable for the outcome.

And this part matters: none of this means AI is useless here. It can genuinely help you get ready — to ask better questions, understand your options in plain words, and organize your notes before you speak to the right person. The rule is simple: AI can help you prepare, organize information, and think of useful questions. Qualified people make consequential decisions and remain accountable for them.

A helpful analogy. It's like using a map app before a hospital visit — it can help you plan the journey, but it doesn't perform the treatment. Use the preparation; leave the decision to the professional.

Your turn. Someone has a serious, worrying medical symptom. What's the responsible use of AI? (a) Ask AI to diagnose it and follow the answer. (b) Use AI to prepare clear questions, then see a qualified doctor. (c) Do nothing and tell no one. Answer: (b). AI helps you prepare; a qualified professional makes the diagnosis and the decision.

Here's the aha: knowing when to step back and involve a person isn't a weakness — it's the most responsible way to use AI.

A practical example. Rosa uses AI to help organize her observations and prepare three questions for the doctor, shares nothing beyond what she needs, and books the appointment. Prepared with AI; decided by a professional. Notice what she does not do: she uses only a generic question-organizing template and enters none of her father's symptoms, identity, records, or medical details into the tool.

A common mistake to avoid: treating a confident AI answer as a substitute for a doctor, a lawyer, or an emergency service. For high-stakes decisions, AI prepares; it never decides.

What you can do today: think of one "a person must decide" situation in your life, and note who the right qualified person or official channel would be.

What you can do now: name three situations where the decision should always involve a qualified person — and say how AI could still help you prepare.

Bringing it together

• •Keep private things private — share the task, not passwords, numbers, medical details, or other people's information. (7.1)
• •Trust the help, then check what matters — confidence is not proof, and checking keeps you in charge. (7.2)
• •Some decisions belong to qualified people — AI can help you prepare; a professional decides and is accountable. (7.3)
• •That's responsible AI: safe habits, human judgment, and you at the center.

Examples

• •Redact first: replace names and numbers with "a client" before asking for help wording a message.
• •The confidence trap: a tidy, sure-sounding fact turns out to be out of date — one quick check catches it.
• •AI helps / not enough: AI drafts a summary; you verify the figures yourself before you send it.
• •Prepare, don't decide: AI helps you write questions for a doctor; the doctor makes the diagnosis.
• •Emergency: in a dangerous situation, you contact emergency services — not a chatbot.

Stories

• •Mei (bookkeeper, Lesson 7.1) — kept a client's details private and still got the help she needed.
• •Idris (community-radio volunteer, Lesson 7.2) — one quick check kept a wrong fact off the air.
• •Rosa (family carer, Lesson 7.3) — used AI to prepare, then let a qualified doctor decide.