ChatGPT feels like it can do anything, which makes the boundaries easy to miss — until something goes wrong. What should you never ask ChatGPT to do? It isn’t just about the requests it refuses; it’s about the requests it accepts and answers badly. I’ve seen people share passwords, upload legal documents, and make medical decisions based on chat answers. Let’s draw the lines clearly.
What Should You Never Ask ChatGPT to Do? The Hard Lines
1. Never share sensitive personal information
Don’t paste passwords, financial account numbers, ID documents, or private health records into any AI chat. Even with data controls set to opt out of training, you’re still sending sensitive data to a third-party server. Screenshots are a sneaky vector here — they often contain account details in the background. If you’re using ChatGPT’s image analysis, crop ruthlessly first.
2. Never treat it as a doctor, lawyer, or financial advisor
ChatGPT can explain medical concepts, outline legal processes, and discuss investment basics — that’s genuinely useful orientation. But it should never be the final word on your health, your lawsuit, or your money. It doesn’t know your full situation, it hallucinates specifics, and it has no professional accountability. Use it to prepare better questions for the actual professional, not to replace them.
3. Never ask for help with wrongdoing
This one’s obvious but worth stating: fraud, hacking, academic dishonesty, creating deceptive content — ChatGPT will refuse much of this, but attempts can get accounts flagged. More importantly, the “it refused so I rephrased” game isn’t clever; it’s just a slower way to do something you shouldn’t do.
4. Never outsource judgment on high-stakes decisions
Hiring, firing, medical triage, legal strategy — AI can organize information around these decisions, but the call has to be human. Models don’t understand context the way you do, they can’t be held accountable, and their confident tone masks real uncertainty. If the decision affects someone’s life or livelihood, AI advises; humans decide.
The Sneaky Ones: Requests It Accepts But Shouldn’t Fully Trust
Beyond the hard lines, there’s a gray zone where ChatGPT complies but you should stay skeptical:
- Precise citations and quotes: frequently invented — always verify
- Real-time critical info: flight statuses, emergency guidance, market data — use authoritative live sources
- Anything requiring taste or ethics: it can simulate judgment but doesn’t have any
- Evaluating people: screening resumes, judging character from text — biased and unreliable
This is where fact-checking ChatGPT’s responses becomes essential rather than optional. The gray zone is where most real-world harm happens — not dramatic refusals, but quiet over-trust.
Privacy Boundaries Most People Miss
A few specific don’ts that trip people up:
- Don’t upload others’ private info — screenshots of conversations, someone’s documents, photos of people — without a real reason
- Don’t paste work documents without checking your employer’s AI policy; many companies prohibit or restrict it
- Don’t assume “delete chat” means deleted everywhere — check the data retention policy for your plan
- Don’t use it for children’s sensitive data — extra caution with anything involving minors
On the file side specifically, it’s worth knowing what file types ChatGPT can analyze — not to upload more, but to understand exactly what’s technically possible so you can set boundaries deliberately. And if you work with visual content, the same caution applies when you use ChatGPT to describe images: don’t upload photos containing other people’s private information.
What You SHOULD Do Instead
The pattern is simple: use ChatGPT for drafting, explaining, organizing, and brainstorming — and keep humans in charge of deciding, verifying, and handling sensitive data. When you’re unsure, ask yourself: “If this answer is wrong, what happens?” If the answer is “nothing much,” proceed. If it’s “someone gets hurt, fired, misdiagnosed, or defrauded,” stop and get a human expert.
Gray Areas: Where Your Judgment Matters Most
Beyond the hard lines, there’s territory where ChatGPT will comply but you should pause:
- Writing about real people: reviews, complaints, character assessments — AI-generated text about identifiable people can defame. Verify everything, consider whether you should write it at all.
- Emotional decisions: “should I quit my job / end this relationship” — ChatGPT can list pros and cons brilliantly, but it doesn’t know your life. Use it for clarity, not verdicts.
- Children’s data: homework help is fine; sharing kids’ personal information, photos, or school records needs extra caution.
- Security advice: it can explain concepts but shouldn’t be your only source for securing real systems — verify against authoritative guides.
The common thread: these are areas where the consequences of error fall on real people, and no disclaimer in a chat window changes that. When in doubt, ask: “If this is wrong, who gets hurt?”
Frequently Asked Questions
Can ChatGPT give medical advice?
It can explain medical concepts, but shouldn’t guide your health decisions. Use it to understand, then see a professional to decide.
What happens if I paste passwords into ChatGPT?
Change them. Immediately. Treat any credential shared with any AI as compromised — then check your data-sharing settings going forward.
Can it help draft legal documents?
It can draft and explain, but legal documents have real consequences — have a lawyer review anything that matters. AI drafts + human expert review is the right combo.
Will ChatGPT refuse illegal requests?
Most of them, yes — and attempts to evade refusals can get accounts flagged. The refusal system isn’t perfect, which is why your own judgment is the real safeguard.
The Takeaway
The lines aren’t complicated: no sensitive data in, no professional advice out as gospel, no wrongdoing, no high-stakes decisions delegated. Everything else is about proportionate skepticism — verify more when stakes are higher. ChatGPT is an extraordinary assistant with no judgment of its own; the judgment has to come from you. Keep that division of labor clear and you’ll avoid nearly every pitfall. For the official word on usage policies, see OpenAI’s site and its published usage guidelines.
