5 Things You Should Never Ask AI to Do

AI tools are remarkable. They can write, research, analyze, generate images, and assist with almost any knowledge work task. But there are specific things you should never ask AI to do — not because AI will refuse, but because doing so puts you, and potentially others, at real risk. This isn’t about fear of AI. It’s about using it responsibly.

Understanding AI’s Limitations and Risks

AI tools are not neutral. They can produce harmful outputs, make confident errors in high-stakes domains, and create legal and ethical exposure if misused. The five things covered in this guide aren’t edge cases — they’re common mistakes that real people make every day, often without realizing the consequences. For a broader look at responsible AI use, see our guides on AI ethics and whether AI is safe.

1. Medical Diagnosis or Treatment Decisions

Every week, millions of people type their symptoms into ChatGPT or other AI tools and ask what’s wrong with them. Sometimes they get useful general health information. Sometimes they get dangerous misinformation delivered with complete confidence. The problem isn’t just that AI can be wrong about medical facts — it’s that AI cannot examine you, cannot see your full medical history, and cannot integrate the countless contextual factors that go into real clinical diagnosis.

Why this is dangerous: A real case: a person typed their symptoms into AI, got a reassuring response suggesting it was likely benign, and delayed seeking care. What they actually had required urgent treatment. AI can’t order blood work, it can’t palpate your abdomen, it doesn’t know your medication history, your allergies, or your family history. Even highly educated AI systems regularly get medical details wrong in ways that could matter enormously.

What to do instead: Use AI for general health education — understanding what a condition is, what questions to ask your doctor, how a medication generally works. Never use it to replace clinical judgment for diagnosis or treatment decisions. If you have symptoms that concern you, see a healthcare professional. The time you’d save isn’t worth the risk. See our AI privacy guide for related data concerns.

The risks here are also compounded by the fact that AI chatbots are designed to be helpful and agreeable — they tend to validate whatever framing you bring to them. If you go in convinced you have a minor condition, AI is likely to support that framing even when the evidence should point elsewhere.

2. Legal Advice for Your Specific Situation

AI can explain legal concepts clearly and generally. It can describe how contracts work, what the elements of a negligence claim are, what rights you generally have as a tenant. This is genuinely useful. What it cannot do is give you reliable legal advice for your specific situation, your specific jurisdiction, your specific document, or your specific dispute.

Why this is dangerous: Law is jurisdictionally specific, factually dependent, and constantly evolving. A provision that’s enforceable in one state is void in another. A legal strategy that works in one set of facts is disastrous in a slightly different set. AI doesn’t know the specific judge assigned to your case, the specific local rules that apply, or the specific precedent that controls in your circuit. And critically, AI has been documented to hallucinate legal cases — making up fake citations that don’t exist, which has caused real harm when lawyers used them in court.

What to do instead: Use AI to educate yourself on legal concepts and prepare better questions for a real attorney. Use it to understand the general landscape of an issue before your consultation, making that consultation more productive. For actual legal decisions — signing contracts, responding to legal demands, making custody arrangements — consult a licensed attorney. Many offer free initial consultations.

This doesn’t mean AI legal tools are useless — some specialized legal AI tools are designed for specific, well-defined tasks and are used by actual lawyers. The risk is in using general-purpose AI as a substitute for professional legal judgment in consequential situations.

3. Major Financial Decisions

Should you invest in this stock? Should you take out a second mortgage? Is this investment opportunity legitimate? Should you roll over your 401(k)? These are questions people regularly ask AI — and the answers they get can sound authoritative, detailed, and completely wrong.

Why this is dangerous: AI does not know your full financial picture — your income, debts, risk tolerance, tax situation, retirement timeline, family obligations, or goals. It cannot access real-time market data reliably. It cannot be held liable if its advice causes you financial harm. And the consequences of bad financial decisions can be irreversible: retirement savings depleted, debt spirals, missed opportunities. AI also cannot detect fraud or spot red flags that an experienced financial professional would recognize immediately.

What to do instead: Use AI to educate yourself on financial concepts — understanding what compound interest means, what the difference between a Roth and Traditional IRA is, how to read a balance sheet. Use it to prepare questions for a financial advisor. For major financial decisions, work with a licensed fiduciary financial advisor who is legally required to act in your interest. See our AI safety guide for more on this.

There’s also a specific scam risk here: AI-generated financial advice can be weaponized to seem more credible. Fake investment schemes now use AI-generated content to appear legitimate. Be especially skeptical of any financial recommendation that comes through an AI interface you didn’t seek out yourself.

---

4. Creating Content About Real, Identifiable People

This one surprises people. AI is great for writing — but using it to create realistic-seeming content about real people (quotes they didn’t say, scenarios they weren’t in, details about their private lives) creates serious legal and ethical problems. This applies to celebrities, public figures, your colleagues, your neighbors, and anyone else.

Why this is dangerous: Defamation: If AI generates false statements of fact about a real person and you publish them, you can be liable for defamation. This applies even if you didn’t intend for the content to be taken as real. Privacy violations: AI can be prompted to generate content that reveals or speculates about private information. Even when you don’t intend harm, distributing this content can violate privacy laws. Deepfakes and synthetic media: AI-generated realistic images, audio, and video of real people are increasingly regulated, and non-consensual intimate imagery laws are being rapidly expanded. What seems like a creative exercise can carry criminal liability.

What to do instead: Keep AI-generated content about real people clearly labeled as fictional, satirical, or speculative. Never generate realistic-seeming quotes, statements, or scenarios for real people without clear framing. Avoid generating any intimate, embarrassing, or private content about identifiable individuals. When in doubt, use fictional characters. The creative possibilities with AI are vast without needing to involve real people’s identities.

This also has professional implications — generating realistic communications in the name of real colleagues or executives creates fraud risk. The line between a “writing exercise” and actionable impersonation can be thinner than people realize. Check our AI mistakes guide for related pitfalls.

5. Sharing Passwords, Financial Credentials, or Sensitive Personal Data

This is the one where people say “I would never do that” — and then do it anyway without realizing it. Sharing sensitive data doesn’t always mean typing your password into a chat. It can mean pasting a work document that contains client names and contact information, uploading a PDF with contract terms that include proprietary details, including your full name, address, and birthdate in a prompt context, or asking AI to help draft a message that includes account numbers or policy details.

Why this is dangerous: Most consumer AI tools are not designed for sensitive data. Your conversations may be used for model training. Data breaches at AI companies have occurred and will occur again. Even if the company’s privacy policy is good, you’re creating a vector for exposure. For business users, there are often regulatory and compliance reasons (HIPAA, GDPR, SOC 2) why certain data should never go into external AI systems.

There’s also a social engineering risk: AI chatbots can be manipulated. Someone could potentially construct a scenario where they use AI to help them access or exploit credentials that were shared in a conversation. The attack surface you create by sharing sensitive data with AI tools is real.

What to do instead: Develop a personal “AI data diet.” Before pasting anything into an AI tool, ask: would I be comfortable if this appeared in a data breach? If not, anonymize it or leave it out. For businesses, establish clear policies about what categories of data can and cannot enter AI tools. Many enterprises use on-premise or enterprise AI deployments specifically to address this risk. See our full AI privacy guide for practical steps.

---

10 More Things to Never Ask AI Without Caveats

• Tax advice for your specific situation. AI can explain general tax concepts; only a licensed CPA or EA can advise on your specific filing situation.
• Therapy or crisis intervention. AI is not a substitute for licensed mental health care. For crisis, contact 988 (Suicide and Crisis Lifeline, US).
• Investment recommendations. AI can explain investment concepts; only a fiduciary financial advisor can advise on portfolio decisions.
• Custody or family law advice. Family law is jurisdiction-specific and life-altering. AI can explain general concepts; a licensed family law attorney is essential for actual decisions.
• Drug interaction analysis without pharmacist review. AI can flag possible interactions; a pharmacist verifies them. The flag is useful; the decision is theirs.
• Code that touches production without review. AI-generated code for production must go through human review. The defaults are not safe enough for live systems.
• Real-time pricing for stocks, crypto, or commodities. AI training data is stale; real-time data requires real-time sources.
• Specific legal advice for ongoing litigation. AI can explain legal concepts; attorneys must handle active cases.
• Medical dosage calculations for specific patients. AI can explain dosing principles; clinicians dose patients.
• Anything where the stakes are higher than your ability to verify. The general principle. Verify what you can verify; defer what you cannot.

Frequently Asked Questions

Can I use AI for general health information at all?

Yes — AI is quite good for understanding health topics, learning about conditions, interpreting medical jargon, and preparing for doctor appointments. The line is between education (fine) and diagnosis or treatment decisions for your specific situation (not fine).

What if I just need basic legal information, not advice for my situation?

General legal education is a good use of AI. Understanding what a lease means, what your rights are as an employee in general, what contract clauses typically do — all of this is valuable. Just be clear that “this is how it generally works” is different from “this is what you should do in your specific case.”

Are there AI tools that ARE safe for sensitive data?

Enterprise-grade AI deployments (Azure OpenAI, Claude for Enterprise, etc.) typically offer contractual data protection and don’t use your data for training. If your organization uses these, check the specific terms. For personal use, the safest approach is to simply not input sensitive data.

How do I explain these risks to colleagues who are using AI unsafely?

Focus on concrete scenarios rather than abstract risks. “What would happen if this document showed up in a data breach?” is more persuasive than “AI might not be secure.” Many people respond to specific, relatable examples of what could go wrong.

Is the risk of AI errors different in different domains?

Yes, significantly. AI errors in creative writing are usually harmless. AI errors in medical, legal, and financial contexts can have serious, sometimes irreversible consequences. The threshold for caution should scale with the stakes of being wrong.

---

Understanding AI’s limits is the first step to using it confidently. Explore AI ethics for beginners for a broader look at responsible use, check our how to use AI guide for what AI CAN help with, and review our 10 AI mistakes beginners make to avoid common pitfalls.

Going Deeper: Advanced Strategies and Practical Applications

Understanding the full scope of this topic requires looking beyond the basics and exploring the nuanced strategies that experienced practitioners rely on every day. Whether you are just starting out or looking to refine your existing approach, the insights covered in this section will help you develop a more robust and effective framework. By taking the time to explore these advanced concepts, you position yourself ahead of the curve and gain a competitive edge that is difficult to achieve through surface-level knowledge alone. The most successful people in this space consistently invest in deepening their understanding, and the payoff in terms of results and efficiency is enormous.

Building a Sustainable Long-Term Approach

One of the most common mistakes beginners make is focusing exclusively on short-term wins while neglecting the foundation needed for lasting success. A sustainable long-term approach means setting up systems, workflows, and habits that continue to deliver value over months and years, not just days or weeks. This involves regular review cycles where you assess what is working, what needs adjustment, and where new opportunities have emerged. It also means staying current with evolving best practices and tools, since the landscape in artificial intelligence and digital business shifts rapidly. Those who build adaptable, iterative frameworks consistently outperform those who rely on static, one-time setups. Treat your strategy as a living document that grows alongside your knowledge and your goals.

Common Pitfalls to Avoid

Even experienced practitioners fall into certain traps that can slow progress or undermine results. One of the most frequent pitfalls is over-complicating a workflow before it has been validated at a simpler scale. Start lean, prove the concept, then layer in additional complexity as needed. Another common mistake is ignoring the human element — technology and automation are powerful, but they work best when paired with clear communication, realistic expectations, and ongoing human oversight. Additionally, many people underestimate the importance of documentation. Keeping clear records of what you have tried, what worked, and what did not saves enormous time when revisiting or scaling a process. Finally, do not neglect community and peer learning. Connecting with others who are working through similar challenges accelerates your growth far more than working in isolation.

• Start simple: Validate your core approach before adding complexity.
• Document everything: Track what works and what does not so you can iterate intelligently.
• Stay updated: Subscribe to reputable sources and revisit your strategy quarterly.
• Leverage community: Join forums, groups, and networks where peers share real experiences.
• Measure outcomes: Use clear metrics so you know when to pivot and when to double down.

Practical Tips for Immediate Implementation

Translating knowledge into action is where most people struggle. The gap between understanding a concept and actually implementing it can feel daunting, but breaking the process into small, manageable steps makes it achievable. Begin by identifying the single most impactful change you can make right now — not the most complex or impressive one, but the one that will deliver tangible results with the least friction. Once that first step is running smoothly, add the next layer. This incremental approach reduces overwhelm, builds momentum, and creates a track record of small wins that keeps you motivated. Remember that consistency beats intensity in the long run. A modest improvement applied consistently over three months will outperform a dramatic overhaul that you abandon after two weeks because it was too difficult to maintain.

Measuring Your Progress and Adjusting Course

Progress without measurement is just activity. To truly know whether your efforts are paying off, you need to define clear, specific metrics before you begin and track them consistently over time. These metrics should be tied directly to the outcomes that matter most to you — whether that is revenue, time saved, audience growth, or skill development. Review your numbers on a regular schedule, whether weekly, biweekly, or monthly, and use what you find to make informed decisions. When a metric is trending in the wrong direction, treat it as useful information rather than a failure. Ask why the number moved, what variables changed, and what you can test to improve it. This analytical mindset transforms every result — good or bad — into an opportunity to learn and optimize.

Scaling What Works

Once you have identified an approach that delivers consistent results, the next challenge is scaling it without losing the qualities that made it effective in the first place. Scaling too quickly can introduce inefficiencies, reduce quality, or overwhelm your current infrastructure. A thoughtful scaling strategy involves gradually increasing volume or scope while monitoring your key metrics closely for any signs of degradation. It also means systematizing the elements of your process that are currently manual or dependent on your personal involvement, so that growth does not require a proportional increase in your time and energy. The goal is to build systems that perform reliably at larger scales, freeing you to focus on strategy, innovation, and the higher-level decisions that drive the most value.

The journey from beginner to confident practitioner is rarely a straight line, but it is absolutely achievable with the right mindset, the right tools, and a commitment to ongoing learning. Every expert you admire started from zero and built their knowledge and skills through consistent effort over time. The information and strategies covered throughout this article give you a strong foundation to build on. Take what resonates, apply it in your own context, and keep refining as you go. The most important step is always the next one — so use what you have learned here and put it into practice today.

You May Also Like

• What Is Artificial Intelligence
• Best AI Tools for Beginners
• How to Use AI
• AI Tools Directory
• Best Free AI Courses