How to Ask AI Medical or Legal Questions Privately
Short answer: remove anything that identifies you (names, dates, places, case or record numbers, rare details), ask through a service that needs no account and doesn’t keep chats, or run a model on your own device, and clear your history afterwards. Use the answer to prepare for a doctor or lawyer, not to replace one. AI can be confidently wrong, and a chat with an AI isn’t protected the way a conversation with your doctor or lawyer is.
Disclosure: written by Obscurify.ai, one of the services mentioned. This page is general information, not medical or legal advice. In an emergency, call your local emergency number (911 in the US). In the US, call or text 988 for a suicide or mental health crisis.
Why are AI chats about health or legal problems risky?
Three reasons, each from a primary source:
- They aren’t privileged. In United States v. Heppner (S.D.N.Y., February 17, 2026), Judge Rakoff ruled that a defendant’s written exchanges with Claude were not protected by attorney-client privilege or the work product doctrine. The court said he had “communicated with a third-party AI platform” whose privacy policy allowed disclosure, so he had no reasonable expectation of confidentiality. The FBI found the chats on devices seized from his home. (Harvard Law Review summary)
- They usually aren’t covered by HIPAA. HIPAA applies to “covered entities and business associates”: health plans, clearinghouses and certain health care providers, and companies working for them. A general chatbot you sign up for yourself is usually neither. (HHS)
- Mainstream chatbots keep and sometimes review chats. Google keeps Gemini chats chosen for human review for up to 3 years and asks users not to enter “confidential information that you wouldn’t want a reviewer to see”. (Google) Claude keeps flagged chats for up to 2 years. (Anthropic) In 2025 a court order made OpenAI preserve ChatGPT conversations, deleted ones included, until September 26, 2025. (OpenAI)
A private service reduces what exists to be stored, reviewed or handed over. It doesn’t make a chat privileged. Only a real lawyer or doctor relationship does that.
How do you ask AI a medical or legal question privately?
- Check it isn’t urgent. Chest pain, trouble breathing, thoughts of self-harm, a court date tomorrow or an arrest: call a professional or emergency line now, not a chatbot.
- Strip identifying details (table below). The model only needs the facts that change the answer.
- Ask without an account where you can. Duck.ai, Brave Leo, Proton Lumo’s guest mode and Obscurify.ai’s guest chat need no sign-up and say they don’t keep chats on their servers. Signed-in ChatGPT, Claude or Gemini tie every question to your account, and Claude also requires a phone number.
- For the most sensitive questions, run a model on your own device. Nothing leaves your computer. Obscurify.ai’s Local Mode runs small models in the browser; desktop apps such as Ollama run larger ones. Small models make more mistakes.
- Hide your IP address if it matters. Use Tor or a VPN you trust. Obscurify.ai works over Tor (ChatGPT over Tor).
- Don’t upload original documents. Lab reports, scans, letters from a court or employer carry names, numbers and file metadata. Retype only the relevant lines, without identifiers.
- Ask for understanding, not a verdict. Good questions: “What does this term mean?”, “What are the usual options for this?”, “What should I ask my doctor or lawyer?”, “What documents should I bring?”
- Check the answer and clear your history. Ask for sources and compare with an official source, such as MedlinePlus for health or your court’s self-help pages for legal questions. Then delete the chat: private services keep history in your browser, and anything on your device can be found if the device is.
What details should you remove?
| Instead of | Write |
|---|---|
| Your name, a relative’s or your doctor’s name | “I”, “my sister”, “my doctor” |
| Date of birth, exact dates | Age range, “about three weeks ago” |
| Street, town, hospital, employer, court | “a mid-sized US state”, “my employer” |
| Case, claim, record, policy or account numbers | Leave them out |
| Rare condition plus rare job plus small town | Keep only what changes the answer |
| Screenshots of portals, letters or scans | The relevant lines, retyped |
| Who did what to whom, in a legal dispute | “Person A”, “Person B”, “the landlord” |
The jurisdiction (country or US state) does change legal answers, so keep that, at the level of a state or country.
What does a private question look like?
Before: “I’m Maria Lopez, 34, at 12 Elm St in Dayton. Dr. Patel at Riverside Clinic prescribed 40 mg of X on Oct 3 and my employer, Acme, says I failed a drug test. Can they fire me? Here’s the letter.”
After: “In Ohio, can an employer act on a positive workplace drug test when the employee has a valid prescription for the medication? What should the employee ask the employer, and what kind of lawyer handles this?”
The second version gets the same useful answer and identifies no one.
Which AI services are best for sensitive questions?
Checked 2026-11-06 on each service’s own pages.
| Option | Account needed | Keeps your chats on its servers? | Notes |
|---|---|---|---|
| Local model on your device | No | Nothing leaves your device | Most private; smaller models, more mistakes |
| Duck.ai | No | No, by default | Free with a daily limit; providers agree to zero data retention |
| Proton Lumo guest | No | No; guest chats erased after the session | Open-source apps; open models |
| Brave Leo | No (in the Brave browser) | No | Free models include Claude Haiku |
| Maple AI | Yes | Stored end-to-end encrypted, unreadable to Maple | Strongest technical guarantee; needs an account |
| Obscurify.ai | No for guest chat; an account is only a random number | No; held in memory up to 1 hour, then deleted | Works over Tor; model companies see the prompt text |
| ChatGPT, Claude, Gemini (signed in) | Yes (Claude also needs a phone number) | Yes, by default | See AI chat that doesn’t store your conversations |
What is AI good and bad at for medical and legal questions?
Useful for: explaining terms in a diagnosis, lab report or contract; listing the usual options so you can ask better questions; summarizing what a type of lawyer or specialist does; drafting a letter that a professional will check.
Not reliable for: diagnosing you, choosing doses, deadlines and filing rules (these differ by court and state), predicting how your case will go, or anything urgent. Models can state wrong facts and invent citations with full confidence, so check anything you’ll act on.
What can’t Obscurify.ai protect?
- The model company sees the text you send, though not who sent it. Text requests only use zero-data-retention routes, which rest on the providers’ published policies; we have no signed agreements.
- No end-to-end encryption or TEE, and our server is closed source. If you need a technical guarantee, Maple AI or a local model is stronger.
- Guest chat uses a standard model (Llama 3.3 70B), which may hedge on frank medical or legal questions. Uncensored models, which answer more plainly, are on the Pro plan (pricing). An uncensored model is more direct, not more correct.
Your prompts and replies aren’t saved on our servers, and an account needs no email or phone. Details are in the privacy policy and on private AI.
Frequently asked questions
Is it safe to ask ChatGPT medical questions?
It’s fine for general understanding, as long as you check what it says with a professional. For privacy, remember that signed-in chats are kept until you delete them, and then for up to 30 days more. Leave out names, dates and record numbers, or use a service that doesn’t keep chats.
Are AI chats covered by HIPAA?
Usually not. HIPAA covers health plans, clearinghouses, certain health care providers and their business associates. A chatbot you sign up for yourself generally isn’t one of those, so HIPAA’s rules don’t apply to what you tell it.
Is what I tell an AI chatbot protected by attorney-client privilege?
No. In United States v. Heppner (2026), a federal court held that a defendant’s chats with Claude were not privileged. Talk to an actual lawyer about anything you might need to keep confidential.
Can my AI chats be used against me in court?
Chats that exist can be requested or seized, whether they’re on the company’s servers or on your own device, as in Heppner. A service that doesn’t store chats, plus clearing your local history, leaves less to find. It doesn’t make anything privileged.
What’s the most private way to ask AI a health question?
A model running on your own device, with no identifying details in the question. Next best: a service with no account that doesn’t keep chats, reached over Tor or a VPN, with your history cleared afterwards.
Should I use an uncensored model for medical or legal questions?
It can help when standard models refuse or hedge on a frank question about drugs, sexual health or criminal law. It won’t be more accurate, and it may be more confident when it’s wrong. Treat its answers the same way: as a starting point to check. See uncensored AI with no email.