It's 11 p.m. Your lab results just landed in the patient portal, and your doctor's office opens in ten hours. So you copy the numbers, paste them into a chatbot, and ask whether a slightly high liver enzyme is something to worry about.
You have company. More than 40 million people ask ChatGPT health questions every day, and more than 5% of all ChatGPT messages worldwide are about health care, according to an OpenAI report covered by Medical Economics. OpenAI also says more than 230 million people ask health and wellness questions on ChatGPT each week, Fortune reported when the company launched ChatGPT Health in January.
The answers can be useful. The bigger question is what happens to everything you typed.
HIPAA Stops at the Chat Window
HIPAA protects health information held by health care providers, health plans, and clearinghouses, plus the vendors that handle data on their behalf, called business associates. A chatbot you open on your own phone is none of those.
The U.S. Department of Health and Human Services (HHS) spells this out. Once health information is sent at a patient's direction to an app that isn't a business associate, it is no longer subject to the protections of the HIPAA Rules, and the provider that sent it isn't liable for what the app does with it next.
That applies to ChatGPT Health, which lets U.S. users connect medical records and apps like Apple Health and MyFitnessPal. OpenAI does not describe it as HIPAA compliant, per Fortune. OpenAI does sell a separate product to hospitals and other provider organizations, ChatGPT for Healthcare, and that one comes with a Business Associate Agreement (BAA). Same brand, different legal rules.
What Does Protect Your Health Chats
Three things, and none of them is HIPAA.
First, the company's own policies. For ChatGPT Health, OpenAI says it won't train its models on personal medical data, excludes health conversations from training by default, and lets users turn on multi-factor authentication and remove connected apps at any time, per Fortune. Those are meaningful protections. They also live in terms of service, which companies update. Anthropic, for example, changed its consumer terms in 2025 so Claude users choose whether their chats train future models. Opting in extends data retention to five years, and opting out keeps it at 30 days, per Anthropic.
Second, a federal breach rule. The Federal Trade Commission's Health Breach Notification Rule covers many health apps that fall outside HIPAA. Updates that took effect in June 2024 count disclosures a consumer didn't authorize as a breach, according to Kirkland & Ellis. The rule requires notification after the fact. It doesn't stop data from leaving.
Third, your own judgment, because the law hasn't caught up. OpenAI CEO Sam Altman said in 2025 that "if you talk to a therapist or a lawyer or a doctor about those problems, there's legal privilege for it," and "we haven't figured that out yet for when you talk to ChatGPT," TechCrunch reported. Litigation makes that concrete. In May 2025, a federal court ordered OpenAI to preserve ChatGPT output logs as part of The New York Times' copyright lawsuit. The judge lifted that requirement going forward in October 2025, but logs saved under the original order remain accessible, Mashable reported.
Every Chatbot Has Fine Print
ChatGPT gets the headlines because it has the most users. The same questions apply to Gemini, Claude, Copilot, and whatever launches next quarter.
Google's Gemini privacy notice says human reviewers, including reviewers from its service providers, read some conversations. Reviewed chats are kept for up to three years, and deleting your activity doesn't remove them. Google's own advice: "Please don't enter confidential information that you wouldn't want a reviewer to see," per Google's Gemini Apps Privacy Hub.
Read the settings page before you read your lab results to any of them. Look for whether chats train the model, how long they're kept, whether humans review them, and whether a temporary or incognito mode exists.
What to Strip Out Before You Paste Your Lab Results
Hospitals and researchers use an HHS standard called the Safe Harbor method to de-identify health data. It lists 18 identifiers to remove. A chatbot question doesn't need the full treatment, but the list makes a solid checklist. Your A1C doesn't need to know your home address.
Before you paste, remove:
Your name, plus the names of relatives or your employer
Your street address, city, and ZIP code
Full dates tied to you, such as birth date, admission date, or test date (the year alone is fine)
Phone numbers, email addresses, and yes, fax numbers
Social Security, medical record, health plan, and account numbers
Photos that show your face, and screenshots of patient portal pages that display your name or record number
Then ask the general version of the question. "What does an ALT of 62 mean for a 45-year-old woman?" gets you most of the answer without handing over your identity. Skip uploading full PDFs of your records, since a complete document carries every identifier at once. And when a chatbot offers a temporary chat mode or a setting to keep conversations out of training, use it.
None of this replaces your doctor. It keeps a late-night question from turning into a record you can't control.
The Same Leak Is Already Inside Your Practice
Patients aren't the only ones pasting. 75% of knowledge workers used AI at work in 2024, and 78% of those users brought their own AI tools, according to Microsoft and LinkedIn's Work Trend Index, which surveyed 31,000 people in 31 countries.
In a medical office, that looks like a staff member pasting patient notes into a personal chatbot account to clean up a referral letter. The practice has now shared protected health information (PHI) with a vendor that has no BAA with it, which is exactly the kind of disclosure HIPAA's vendor rules exist to prevent. We break down those obligations in our guide to BAAs and vendor risk, and we applied the same test to AI documentation tools in Your AI Scribe Is a Business Associate.
The fix starts with an approved tool, a signed agreement, and a written policy staff will follow. Our HIPAA-grounded review of Claude for Small Business shows how to evaluate a specific tool, and our 90-day AI governance plan lays out the rollout. Financial firms have the same exposure with client data, and the SEC is asking about it, as we covered in The SEC Wants to See Your AI Governance Policy.
Techvera's Managed AI team helps healthcare practices and financial firms choose AI tools that fit their compliance obligations, put the right agreements in place, and train staff on what stays out of the chat window. Check where your practice stands with our Compliance Readiness Checker, or schedule a consultation.
Frequently Asked Questions
Is ChatGPT HIPAA compliant?
Consumer ChatGPT, including ChatGPT Health, is not described by OpenAI as HIPAA compliant, per Fortune. OpenAI's separate ChatGPT for Healthcare product for provider organizations comes with a Business Associate Agreement to support HIPAA-compliant use.
Is it safe to upload my medical records to an AI chatbot?
Once you send health records to a consumer app at your own direction, HIPAA no longer protects them, according to HHS. Your protection depends on the company's privacy policy and settings. Remove identifiers such as your name, address, full dates, and record numbers before you share anything.
Can a medical practice use ChatGPT or other AI chatbots with patient information?
Only through a tool covered by a Business Associate Agreement, with a written policy governing its use. Staff should never paste patient information into personal or consumer chatbot accounts. Healthcare-specific offerings such as ChatGPT for Healthcare include a BAA.
About the Author
Team Techvera
Techvera Team
Articles written collaboratively by the Techvera team, combining expertise across cybersecurity, managed services, and digital transformation.