AI chatbots feel like confidants — always available, never judgmental, endlessly patient. That feeling is the trap. There are things you should never ask an AI chatbot, not because the technology is evil, but because the chat window is the wrong place for them: wrong on privacy, wrong on reliability, wrong on consequences. Here are the seven.

What Should I Never Ask an AI Chatbot? Passwords and Credentials

Never paste passwords, API keys, or login credentials — even when debugging. Chat logs may be stored, reviewed, or used for training. A credential in a chat log is a credential compromised. This is rule zero of sharing personal info with chatbots.

2. Other People’s Private Information

Client data, colleagues’ personal details, your kids’ information, someone’s medical or financial specifics. You have no right to volunteer their privacy to a system you don’t control. Anonymize or leave it out — always.

3. Questions Where a Wrong Answer Could Hurt Someone

Medical diagnoses, drug dosages, legal strategy for your actual case, safety-critical instructions (electrical work, childcare emergencies). The chatbot’s confident tone doesn’t change its nature: a pattern-matcher, not a professional. The risks of AI medical advice spell out how this goes wrong. For high-stakes questions, the chatbot is a starting point for research, never the answer.

4. Confidential Work Material

Unreleased product plans, proprietary code, internal strategy docs, merger details. Many companies now explicitly forbid this — check your employer’s AI policy before pasting anything work-related. The FTC has also warned businesses about the data security implications of employees feeding sensitive info into AI tools.

5. Financial Account Details

Full account numbers, tax documents with identifiers, complete financial statements. If you want AI budgeting help, use anonymized summaries — see managing your budget with AI for the safe method. The insight doesn’t require the identifiers.

6. Anything You’d Be Devastated to See Quoted

Deeply personal confessions, relationship details, embarrassing questions — the training-data question looms over all of it. Assume anything you type could theoretically resurface in some form. The chatbot is not your therapist, your diary, or your priest. It’s software run by a company.

7. “Is This True?” About High-Stakes Claims

Using the chatbot as your sole fact-checker for consequential claims — a news event, a legal right, a health scare — is asking the wrong tool. Chatbots hallucinate; their accuracy is uneven by design. Verify against primary sources, always. This is also why chatbots aren’t search engines: they generate, they don’t retrieve.

The Safe-Use Cheat Sheet (Save This)

Tape this to your monitor — the quick-reference version of everything above.

GREEN — Go ahead: Brainstorming and ideation. Drafting (emails, posts, plans). Explaining concepts. Summarizing documents. Learning and tutoring. Coding help (reviewed). Creative writing assistance. Math and analysis (verified). Language translation (nuance-checked).
Rule: AI proposes, you dispose. Review everything before it matters.

YELLOW — Proceed with care: Work documents (check employer policy first). Health questions (general info only, anonymized, verified against authoritative sources). Financial planning (anonymized numbers, verified advice). Legal questions (general info, not your case — consult an attorney). News and current events (cross-check; training cutoffs apply).
Rule: useful starting point, never the final word. Verify before acting.

RED — Never: Passwords, credentials, access keys. Other people’s private data. Confidential business information (without authorization). Full financial account details. Medical diagnoses or treatment decisions. Anything you’d be devastated to see quoted. Using AI as your sole source for high-stakes facts.
Rule: the chat window is a postcard, not a diary. If it’s red, it doesn’t go in.

The 10-second test: Before hitting send on anything sensitive, ask: “If this chat log leaked tomorrow, what’s the worst outcome?” If the answer involves your job, your money, your health privacy, or someone else’s trust — don’t send it. This single habit prevents 99% of AI privacy and safety mistakes. Everything else is optimization; this is the foundation. Green freely, yellow carefully, red never. That’s the whole system on an index card.

AI Safety FAQs

What if I already pasted something sensitive?
Don’t panic — assess and act. Delete the conversation/thread immediately. If it was credentials: change them now (assume compromised). If it was work-confidential: inform your manager or IT per company policy — they’ll prefer hearing it from you. If personal: monitor relevant accounts. Then learn the habit going forward; one mistake is education, repeated is negligence.

Are some AI tools safer than others?
Yes — business/enterprise tiers with contractual no-training commitments are meaningfully safer than consumer free tiers. Self-hosted open models are safest for the truly sensitive (data never leaves your machine). Match the tool’s privacy posture to the data’s sensitivity; don’t use a consumer chatbot for what needs an enterprise tool.

How do I teach my kids AI safety?
Three rules they can remember: never share personal info (name, school, address, photos), never believe everything it says (it’s confident, not always right), and tell a parent about anything weird or uncomfortable. Model it yourself — kids adopt the caution they see. The [[i:101|privacy checklist]] applies doubly for young users.

The Principle Behind the List

Every item above follows one rule: match the tool to the stakes. Low-stakes brainstorming, drafting, explaining, exploring — chatbots are wonderful. High-stakes privacy, safety, money, health, other people’s data — wrong venue. It’s the same judgment as voice assistant privacy: not fear, just appropriate caution. Use AI enthusiastically for what it’s good at. Keep the seven categories out of the chat window, and you’ll get all of the benefit with none of the regret.

Teaching Others the System

Once you’ve internalized green-yellow-red, share it — with your team, your family, your less-technical friends. AI safety knowledge is most valuable in the hands of people who’d never read a guide like this: the colleague pasting customer data into a chatbot, the parent sharing kids’ photos with AI apps, the friend asking medical questions at 2am. A thirty-second “hey, quick thing about AI safety” conversation can prevent someone else’s worst mistake. The safest AI era is the one where everyone knows the basics.