Picture this. It is 1 a.m., you cannot sleep, and a chatbot is the only “person” awake. You type out the whole story. The money worry, the fight with a relative, the medical scare, maybe even the thing you did last year that you have never told anyone.
The reply is kind, fast and thoughtful. It feels like a private talk.
It is not.
A chatbot is a product run by a company, and your words travel through that company’s systems. In February 2026, a federal judge in New York went further and ruled that a man’s chats with a public AI chatbot did not get the legal protection people assume a “confidential” conversation gets. We will unpack that ruling in a moment.
This guide is not here to scare you away from AI. I use these tools too. The goal is simple: help you know which things never belong in the chat box, and what to type instead.
What You Will Find in This Guide?
- Passwords, PINs and recovery codes
- ID numbers, passport details and your home address
- Bank, card and tax details
- Medical records and health details
- Legal trouble and your defence plans
- Confidential work files and source code
- Other people’s private details
- Photos and files with hidden data
- Your deepest emotional crises
- Unpublished ideas, business plans and creative work
- Anything you would not want made public
- Your live location and daily routine
- Anything you expect “delete” to erase for good
Why an AI Chat Is Not Like Talking to a Doctor or Lawyer?
When you talk to a doctor or a lawyer, the law and professional ethics protect what you say. A chatbot has neither of those rules behind it.
According to ESET’s October 2026 guide on what to never share with an AI chatbot, major chatbots usually store your inputs. That stored data may be searchable by you later, kept as “memory,” reviewed by the company for safety checks, or used to improve the service. ESET also points out that not every prompt is used for training, and that practices differ between platforms and account settings.
That last part matters. Some apps are stricter than others, and paid or business plans often have stronger protections. But the safest habit is the same everywhere: treat the chat box like a public forum, not a diary.
Even “private” or “temporary” chat modes are not a magic shield. They can keep a chat out of your visible history, but the company may still hold it for a short time to catch abuse or meet legal duties.
Now let us go through the list.
1. Passwords, PINs and Recovery Codes
This one sounds obvious, yet it is the mistake people make when they are in a hurry. Someone pastes a block of text to “fix my login problem” and a password or a backup code is sitting inside it.
Why it is risky?
Chatbot memory and review systems can store what you type. If your login details sit in a saved chat, they sit in one more place that can be exposed. Security recovery codes and answers to security questions count as passwords too.
Do this instead
Use a password manager for storage. If you need help with a login problem, describe the error message and leave the real credentials out. If you ever typed a password into a chat by mistake, change it right away and delete the chat.
2. ID Numbers, Passport Details and Your Home Address
Your full name plus your address plus an ID number is a ready-made kit for identity theft. Chatbots are not built to treat that kit as sensitive the way a bank or government office does.
Why it is risky?
ESET describes this group as sensitive personally identifiable information, and it includes ID and passport details, home addresses and financial details. The company explains that these details are usually heavily protected when you hand them to a regulated provider, but a chatbot is a different story.
Scammers love this kind of detail. The more real facts a fraudster has about you, the more believable the fake call, email or deepfake video becomes.
Quick detour: If you want to see how one leaked detail can become a convincing fake message or voice, our guide to AI deepfake scams shows how these tricks are built and how to spot them.
Do this instead
Use placeholders. Write “[NAME]” or “[CITY]” when you want help with a form, a letter or a résumé. You can fill in the real details yourself after you copy the draft out.
3. Bank, Card and Tax Details
Money details are the next big one. Security company Norton’s guide on what not to share with chatbots lists credit card numbers and bank account details as information to keep out of chats, and it warns that sharing them can lead to financial loss, identity theft and unauthorized transactions. A leaked card number or bank detail can turn into fraud fast.
Why it is risky?
A card number, account number, online banking login or tax ID has one purpose in the wrong hands, and it is not a good one. Chat apps also now connect to other tools, which means your text can travel further than the chat window.
If you let AI tools get anywhere near your money, read our guide on how AI agents and bank accounts interact and how to protect your money before you connect anything.
Do this instead
Ask money questions in general terms. Instead of pasting a statement, say “I earn a salary and have two credit cards, how should I plan to pay down debt?” You get a useful answer without handing over a single real number.
4. Medical Records and Health Details
People now ask chatbots about symptoms, test results and medicines every day. It feels safe because the chatbot sounds calm and informed.
Why it is risky?
ESET makes a simple point here: an AI is not a healthcare provider. A doctor is bound by ethics and by law to protect your data, while a chatbot is not in the same position. So your test results or treatment details become data in a consumer app, not a protected medical record.
There is also a safety issue beyond privacy. A chatbot can sound sure and still be wrong about health.
We covered one real example in our report on AI medical advice gone wrong, and it shows why health answers need a real professional in the loop.
Do this instead
Ask general questions, such as “What are common reasons for a mild headache that lasts three days?” Keep your name, your lab values and your full medical history out of it. For anything serious or urgent, talk to a doctor.
5. Legal Trouble and Your Defense Plans
This is the item that gives this article its title, and it is the one most people have never heard about.
What the court said?
In February 2026, Judge Jed Rakoff of the Southern District of New York ruled in a case called United States v. Heppner. Orrick, a major law firm, explains the ruling this way: the defendant was a senior executive charged with securities fraud, and he used a public AI assistant (Anthropic’s Claude) to think through his legal situation and defence ideas.
As The Next Web reported, federal agents searched his home and took about 31 documents that recorded those AI conversations. He argued they were protected. The judge disagreed. The ruling said the chats were not covered by attorney-client privilege or by the work product doctrine, mainly because an AI chatbot is not a lawyer and the company behind it owes no duty of confidentiality.
Why it is not the whole story?
Be fair about what this ruling is. Lawyers at Baker Botts, writing on Mondaq, call it a narrow decision: one defendant, acting alone, on a consumer AI tool, without his lawyer’s involvement.
It is also not the only view out there. Around the same time, a federal magistrate judge in Michigan went the other way on work product protection for a person representing themselves, so the law is still being sorted out.
The safe takeaway is not “AI always destroys privilege.” It is this: do not assume your AI chat is protected the way a talk with your lawyer is.
Do this instead
If you are in a legal problem, talk to a licensed lawyer first. Use AI for general education only, like “What does a statute of limitations mean?” Do not type your story, names, dates or your plan. And never use a chatbot to confess something you would not want read aloud in a courtroom.
6. Confidential Work Files and Source Code
Chatbots are great at summarizing a long report or checking a block of code. That is exactly why people paste in things they should not.
Why it is risky?
Tech.co’s list of things never to share with ChatGPT puts sensitive company data at the top, and it warns that opting out of AI training is not always easy or even possible.
The best-known example is from 2023, when Samsung employees reportedly pasted proprietary source code and internal meeting notes into a chatbot while trying to get work done faster. A shortcut that saved a few minutes put company secrets outside the company’s control.
Lawyers have also raised a quieter risk. If you paste a trade secret into a consumer tool under broad terms, you may weaken your own claim that you kept it secret.
If you are curious how much further this goes once tools start acting on your behalf, our guide on things you must never hand over to an AI agent is a smart next read.
Do this instead
Check your company’s AI policy first. If your employer provides an approved business tool, use that one for work files. If not, summarize the task yourself in general terms and strip out client names, product plans, prices and code that is not public.
7. Other People’s Private Details
You might be careful with your own information and still hand over someone else’s without thinking. A coworker’s performance issue, a customer list, a friend’s health news, a family member’s address. All of it feels like “just context” when you type it.
Why it is risky?
The person you are talking about never agreed to that. They cannot ask the company to remove it, and they may not even know the chat exists. If the data ever leaks or gets reviewed, they carry the risk, not you.
For work, it can also break privacy rules or contracts. Customer records are not yours to paste into a public tool.
Privacy for the people around us gets even more complicated with wearable tech, and our report on Meta Muse AI glasses and privacy looks at that side of the story.
Do this instead
Swap names for roles. Write “my manager” or “a customer” instead of real names, and leave out phone numbers, emails and addresses. If the advice still works without the identifying details, you did it right.
8. Photos and Files With Hidden Data
A chat is not only the words you type. Every photo, screenshot, PDF and voice note you upload is part of the conversation too.
Why it is risky?
The ESET guide mentioned earlier points out that files can carry sensitive details you never meant to share, hidden in metadata, comments or background content. A phone photo can hold location and time information. A Word document can hold old comments and tracked edits you forgot about. A screenshot can show a notification, a name or an account number in the corner.
A chatbot can read all of that, and it may use what it finds to answer you. That is helpful when it works for you, and a problem when it exposes what you wanted hidden.
Do this instead
Before you upload, look closely. Crop the image, hide notifications, turn off location tagging in your camera settings, and open documents to remove comments and hidden text. For anything private, a plain copy of the text you actually need is safer than the whole file.
9. Your Deepest Emotional Crises
This one is hard to write, because I understand why people do it. A chatbot never sleeps, never judges and never gets tired of you.
Why it is risky?
OpenAI’s CEO Sam Altman has said it out loud. In a podcast interview covered by TechCrunch, he explained that people, especially young people, use ChatGPT like a therapist or life coach, yet the law gives those chats no confidentiality like a real therapist, doctor or lawyer would provide. He added that it makes sense to want that clarity before using it heavily.
There is a second problem. A chatbot can sound caring and still miss what a trained person would catch. It is a tool, not a counselor.
Teens use these tools more than most people, and our ChatGPT teen safety checklist for parents covers what families should look at tonight.
Do this instead
Use AI for simple things, like organizing your thoughts or writing down questions for a real appointment. For the heavy stuff, reach out to a trusted friend, a family member or a licensed counselor. If you ever feel unsafe or in crisis, contact your local emergency number or a crisis helpline in your country right away.
Quick Swap Table: Say This, Not That
| Instead of typing this | Type this |
|---|---|
| “Here is my full statement with account numbers” | “I have two credit cards and a student loan. How do I plan payoff?” |
| “My boss [real name] said X about client [real name]” | “My manager criticized my report in front of a client. How should I respond?” |
| “Here is my passport scan” | “Help me write a polite email asking for a visa appointment.” |
| “Here is my lab report” | “What does a high cholesterol reading generally mean?” |
| “Here is my company’s Q3 plan file” | “Give me a template for a quarterly plan summary.” |
10. Unpublished Ideas, Business Plans and Creative Work
If you are a writer, designer, founder or student, this one is for you. A half-finished novel, a startup pitch, a song you have not released, a product idea you are still testing. It all feels safe in a chat box.
Why it is risky?
Chatbots store what you type, and some use inputs to improve their services depending on the platform and your settings. Once your unreleased work sits in a third-party system, you have less control over where it travels. Experts also list creative works and intellectual property among the things to keep out of public AI tools.
There is a practical side too. If you later need to prove an idea was yours and confidential, a trail of pasted drafts in a consumer chat app does not help your case.
Creators who want to understand where this is all heading can read our deep dive on AI in Hollywood and what it means for writers and artists before sharing any more work.
Do this instead
Ask for help with structure, not substance. Say “What is a good outline for a three-act thriller?” rather than pasting your full manuscript. For business plans, share a simplified version with no names, numbers or secret sauce.
11. Anything You Would Not Want Made Public
Here is a simple rule I now follow: if I would not post it on a public forum, I do not type it into a chatbot.
Why it is risky?
Chats can end up in places people never expected. In 2025, thousands of conversations that users shared through public links showed up in search results. Those chats were not hacked from private accounts. People had simply used a sharing feature without realizing how public it could become. Yahoo Tech also reported that a trove of Claude conversations appeared in Google search results, which shows this is not a problem for just one company.
Third parties can be a risk as well. The Register reported on data brokers selling deeply personal bot transcripts, including a Koi Security report about browser extensions that claimed to protect privacy while collecting conversations from millions of users.
Do this instead
Think before you press “share link.” Check your shared links list regularly and delete the ones you do not need. Be picky about browser extensions that say they work with AI chats, and install only ones you truly trust.
12. Your Live Location and Daily Routine
“I am home alone this week.” “I leave for work at 7 every morning.” “I am flying out Friday and the house will be empty.”
Those lines look harmless in a chat about travel tips or schedules. Put together, they paint a clear picture of your life.
Why it is risky?
Chatbots already pick up more about you than most people expect. Kim Komando notes that a chatbot can gather information about where you live from your IP address, along with other permissions you accepted in the terms. If you also enable memory or connect apps, your routine can become even easier to piece together.
A detailed routine is useful to the wrong person, whether that is a scammer, a stalker or a thief.
Do this instead
Ask for general plans. “Plan a 5-day trip to Goa on a medium budget” works fine without telling the bot exactly when your home will be empty. Skip your exact address, school or workplace name, and your daily schedule.
13. Anything You Expect “Delete” to Erase for Good
The last item is a habit more than a topic. Many people type something sensitive, regret it, and tap delete. Then they relax.
Why it is risky?
Deleting a chat removes it from your view, but that is not always the same as erasing every copy. Security experts say that deleting a conversation should not be treated as proof that all traces are gone, since retention periods and exceptions vary between services.
Business Insider reported that most deleted ChatGPT conversations are permanently wiped after 30 days, and also that Altman warned therapy-style chats might not stay private in a lawsuit. So the real rule is simple: prevention beats cleanup.
Do this instead
Do not type it in the first place. If you already did, change any exposed passwords right away, delete the chat, clear any saved memory, and tell your employer’s security team if work data was involved.
Your 8-Step Safety Checklist for Using AI Chatbots
Keep this list close. It takes about five minutes to set up and protects you every day.
- Read the privacy settings. Bitdefender’s chatbot privacy tips recommend reading the policy and turning off data use for training where that option exists.
- Use temporary or incognito chat for sensitive questions, but do not treat it as a vault.
- Review and clear chatbot memory every few weeks.
- Replace names, numbers and addresses with placeholders.
- Check and delete old shared links.
- Use only approved business tools for work files.
- Keep a password manager, so passwords never need to be typed into a chat.
- Ask yourself one question before you hit send: “Would I be fine if this appeared in a data breach?”
What Is Changing: The Law Is Catching Up
Lawmakers have noticed the gap. According to a Future of Privacy Forum tracker reported by Local News 8, almost 100 bills aimed at regulating AI chatbots were introduced in US state legislatures in 2026 alone.
Some of the pressure comes from child safety concerns, too.
If you want the background on that fight, our report on state attorneys general suing OpenAI over ChatGPT and kids explains what parents need to know.
Until clear rules arrive, your habits are your best protection.
Frequently Asked Questions
Are AI chatbot conversations private?
Not necessarily. How chats are stored, reviewed or used depends on the company, the product, your account type and your settings. Treat a consumer chatbot as a third-party service, not a private confidant.
Can courts get my AI chats?
They can in some situations. A 2026 federal ruling in New York found that a defendant’s chats with a public AI tool were not protected by attorney-client privilege, and Minnesota Lawyer covered how the decision may affect clients. It was a narrow decision, and other courts may see it differently, so the safest plan is to avoid typing legal secrets into a chatbot.
Does temporary or incognito chat make AI safe?
It helps, but it is not a guarantee. These modes may keep a chat out of your history or out of training, yet the company can still hold the data briefly for safety or legal reasons. They also do not cover files you send through connected apps or links you share.
Are paid or business plans safer?
Often, yes. Business plans can offer stronger contracts, admin controls and limits on training use. You still need to follow your workplace rules and share only what the tool is approved to handle.
What should I do if I already shared something sensitive?
Act based on what you shared. Change passwords and recovery codes first, then delete the chat and clear saved memory. If it involved work or customer data, report it to your employer.
Is it still safe to use AI chatbots at all?
Yes, for the right tasks. Brainstorming, drafting, learning and general questions are fine. The goal is not to stop using AI. The goal is to stop treating it like a locked diary.
Final Thoughts
Go back to that 1 a.m. moment from the start. The urge to type everything out is human, and a chatbot will happily listen. But a chatbot is not a lawyer, a doctor or a therapist, and the law does not treat it like one.
So keep it simple. Share the question, not the identity. Share the situation, not the secrets. Ask yourself the public forum test before every send.
Use AI as a powerful assistant, not as your safe. That one habit covers most of the 13 items above.
Related reads from oversmartai.com
- Learn how an AI agent got caught up in a case in our OpenAI agent Medicare hack report.
- See how attackers are using AI for hacking in our story on the Skynet AI agent and autonomous cyberattack tool.
- Read the full list of AI deepfake scandals to see how personal details get misused.
- Check the hidden dangers of letting AI run your social media before you automate your accounts.
This article is for general information and is not legal, medical or financial advice.