Is ChatGPT HIPAA Compliant? Free, Business and Clinician Plans

ChatGPT is not HIPAA compliant on its own, and it depends on the plan. Free, Go, Plus, Pro and ChatGPT Business cannot sign a BAA, so no patient data goes in. A BAA is available for ChatGPT for Clinicians (currently free for individually verified clinicians), ChatGPT for Healthcare, and sales-managed Enterprise and Edu workspaces.

The gap that catches small practices is Business. It looks like the grown-up plan: admin controls, no training on your content, shared workspace. Yet OpenAI says it does not offer a BAA for it. A practice that pays for Business and assumes it is covered has a worse problem than one on the free plan, because the paying team feels safe. For most small practices the real choice is narrower than the plan list suggests: keep patient data out of ChatGPT altogether, or let each eligible clinician use their own covered Clinicians workspace while everyone else sticks to non-patient work.

Follow me on Instagram@sagnikteaches

ChatGPT plans and BAA availability, row by row

The table reflects OpenAI's help articles on BAAs, HIPAA-eligible products, ChatGPT for Clinicians, ChatGPT for Healthcare and Health in ChatGPT, as of November 2026. Product names, plans and eligibility change, so confirm against the live pages before you rely on any row.

Connect on LinkedInSagnik Bhattacharya

Two cautions. First, "BAA available" never means "BAA automatic". Someone with authority to sign has to review and accept the agreement for the right workspace, and the product settings OpenAI prescribes have to be followed. Second, even a covered plan does not cover every feature. OpenAI's list of HIPAA-eligible functionality excludes improved memory, event-triggered scheduled tasks, internet access in canvas and code blocks, and browser or network access for Work in the cloud, and it says a feature not on the covered list is not covered. The Clinicians article adds that no patient details belong in queries to the public healthcare data apps (PubMed, ClinicalTrials.gov and similar) or in third-party GPTs and connected apps without their own agreements. A BAA covers the services it names, nothing more.

Subscribe on YouTube@sagnikteaches

One detail sets ChatGPT apart from some other assistants. OpenAI says HIPAA-eligible workspaces run web search and deep research against its own search index and do not send queries to third-party search providers such as Bing. Compare that with how Microsoft 365 Copilot treats web search queries, where the generated queries sit outside the BAA.

Why Business has no BAA and what that means for you

ChatGPT Business is priced and designed for teams: it offers admin controls and, by default, does not train on your business content. Those are privacy features. HIPAA asks a narrower question: has the vendor agreed in writing to the obligations of a business associate for this service? For Business, OpenAI's answer is no. So the practical rule is the same as for the free plan. If the text would identify a patient, it does not belong in Business.

That does not make Business useless to a clinic. It makes it a tool for everything that is not patient data: rewriting a recall letter template, drafting the practice's own policies, summarising a published guideline, planning a staff rota. Our guide to what a small practice needs first for AI sorts those jobs from the patient-facing ones.

The one route open to a small practice: ChatGPT for Clinicians

OpenAI points individual clinicians to ChatGPT for Clinicians. According to OpenAI's help article, eligibility currently covers physicians (MD or DO), nurse practitioners, physician assistants and pharmacists, with a valid NPI and a licence that can be verified through a third-party provider. It is free at launch for verified clinicians. Sign-up creates a separate Clinicians workspace you reach from the account switcher, and your existing ChatGPT workspace stays as it was. An eligible clinician can review and sign a BAA for that workspace under Settings, then Agreements. OpenAI says that where you need a BAA for multiple users, central controls or an organisation-wide deployment, you should look at ChatGPT for Healthcare instead.

Three consequences for a small practice:

• It is personal. The BAA is signed by the individual for their workspace. A receptionist, a hygienist or a practice manager is not covered by a doctor's account.
• It is narrower than the practice's work. Not every clinical role is on the eligible list today, so a chiropractor, a physiotherapist or a counsellor may not qualify. Check the current list rather than assuming.
• Features are limited. The exclusions on OpenAI's covered-functionality list (improved memory, event-triggered scheduled tasks, third-party GPTs and apps without their own agreements) still apply, image generation is not offered, and the Epic plugin is not available to individual Clinicians accounts. Avoid anything the list does not name as covered.

For a deeper look at what the clinician BAA does and does not cover, see ChatGPT for Clinicians: does the individual BAA cover your practice?

Sorting the jobs in a six-person family practice

Say a family practice has two doctors, one nurse practitioner, a practice manager and two front-desk staff. Everyone has a ChatGPT login of some kind, and nobody has written down what may go in. A sensible sort might look like this.

The practical outcome: the two doctors and the nurse practitioner each decide whether to take up ChatGPT for Clinicians and, if eligible, sign their own BAA. Everyone else, including the practice manager, uses ChatGPT only for the non-patient rows. That one page of rules does more for compliance than any plan upgrade.

The money side is small. The three Clinicians workspaces cost nothing at launch. If the practice also wants a shared Business workspace for the non-patient rows, six Standard seats at $20 a seat a month on annual billing come to $120 a month, or $150 at the $25 monthly rate. Two of the six rows (the referral letter and the portal replies) are the ones that need a BAA, and in this practice only the three clinicians can do them in ChatGPT at all.

Do-not-paste rules for staff

Put these on one sheet, show them at a team meeting and ask each person to confirm they have read them. Adapt the wording to your practice.

AI tool rules for [practice name]

1. Never type or paste a patient's name, date of birth, address, phone, email, insurance or record number, photo, or appointment date into any AI tool unless the Practice Manager has confirmed it is a covered workspace under a signed BAA.
2. "Covered" means: the tool, the plan and the feature are on the practice's approved list. A plan being paid for is not the same as covered.
3. Removing the name is not enough. Ages in the 90s, rare conditions, small towns, exact dates and job titles can still identify someone.
4. Do not upload scans, referral letters, lab results, call recordings or screenshots of the patient record.
5. Do not use personal accounts for work content.
6. Memory, connected apps and scheduled tasks stay switched off unless the approved list says otherwise.
7. If you paste something by mistake, tell [privacy lead] the same day. Nobody is in trouble for reporting.
8. AI drafts are checked by a clinician or the manager before anything is sent or filed.

The list is deliberately short. Rule 2 does most of the work, because it stops the assumption that "we pay for Business" means "we're covered".

Before and after: a message that should not be pasted

Here is a realistic mistake and the safe version. The first is what a busy receptionist might paste into a free chat.

Reply to this patient message. "Hi, it's Maria Gomez, born 4 June 1961. My results from the 12th show high cholesterol. Can I move to a statin? Phone 555 0142."

That has a name, a date of birth, a test date, a health condition and a phone number. None of it belongs in a plan with no BAA. The safe version removes the identity and asks for the shape of the reply only:

Write a short, warm reply template for a patient who asks whether to start medication after a blood test result. Say the clinician will review the result, we will call within two working days, and the patient should not change any medication until then. Do not give medical advice.

The template then goes into the record or the portal and the receptionist adds the patient's name by hand. It takes thirty seconds longer and keeps the identifiers where they belong.

De-identifying a note properly: what the 18 identifiers mean

HIPAA's de-identification route (the Safe Harbor method) asks you to remove 18 types of identifier from health information, and to have no actual knowledge that what remains could identify the person. The list goes well beyond names. It includes all geographic units smaller than a state, every date element except the year (birth, admission, discharge, death and any other date tied to the person), phone and fax numbers, email addresses, record and account numbers, device serial numbers, web addresses, IP addresses, full-face photographs and any other unique code. Ages over 89 must be grouped into a single "90 or older" category.

Here is an illustrative before and after. The "before" line looks harmless to a rushed reader; the "after" is what you would need to reach.

Before: "Mrs J, 93, from a small village near the coast, seen on 14 March with a fall at the care home on Elm Road. Daughter Anna called twice."

After: "Patient aged 90 or older, seen with a fall in a residential setting. Family member called twice."

Notice what had to go: the initial, the exact age, the place, the date, the street and the daughter's name. If you cannot be sure you removed everything, assume the text is still identifiable. For a busy reception desk, a simpler rule works better than teaching the 18-item list: nothing that came out of the patient record goes into a tool that is not on the approved list.

An approved-tools list, filled in

The rules sheet above refers to an approved list. Keep it on one page, owned by one named person, and reviewed on a fixed date. A filled-in example for the family practice might read like this.

The dates and names are invented for illustration. The structure matters more: each row states the plan, whether a BAA exists, what the tool is allowed to do, which features are off, and when someone last looked.

Questions to put to OpenAI before you sign

If you are an organisation considering ChatGPT for Healthcare or an Enterprise workspace, treat the first conversation as due diligence. Write the questions down and keep the answers with the BAA.

1. Which features and connected apps are covered by the BAA, and which are excluded? Ask for the list in writing.
2. How long are prompts, files and outputs retained, and how does an admin delete them?
3. Can the administrator see usage by user, and export it for an audit?
4. What happens to a departing employee's chats, and who owns them?
5. What is the process and the time frame for notifying us of a security incident?
6. OpenAI lists voice, ChatGPT Record and Work on web, mobile and desktop as covered: which of those will our staff use, and how will we hear when the covered list changes?
7. Which sub-processors handle our data?

Answers that are vague, or "we will check", are information in themselves. A covered tool should be able to say plainly what is in and out of scope.

A short incident script for a pasted-in mistake

Every practice that uses AI will eventually have someone paste the wrong thing. Decide now what happens next, so it is not invented under stress. A one-paragraph procedure is enough.

If you think you have put patient information into a tool that is not on the approved list:
1. Stop. Do not continue the chat.
2. Tell [privacy lead] straight away. Say which tool, which account, roughly what was pasted and when.
3. Do not try to fix it quietly. Deleting the chat may help, but only after the lead has noted what it contained.
4. The lead decides, with the practice's legal or compliance adviser where needed, whether the event needs to be treated as a reportable incident, and records the decision either way.
5. We look at why it happened and change the process, not the person.

Step 3 trips people up. Deleting a chat can feel like undoing the mistake, but it also destroys the record you need to judge what happened. Record first, then delete.

Roll it out in four short steps

Week 1: find out what is happening. Run the five-minute check above and write down the answers. Week 2: decide the lines. Fill in the approved-tools list and agree the rules sheet with the clinicians and the manager. Week 3: sign what needs signing. Any clinician who is eligible and wants to use a covered workspace reviews and signs their own BAA, with your privacy lead checking the settings guide. Week 4: tell the team and test. Run a ten-minute session using the before-and-after examples, then ask each person for one example of something they would like AI to do, and sort it into the right column. Put a review date in the diary for three months later, because plans and eligibility change.

Small practices tend to skip the fourth week. It is the one that matters, because rules nobody has practised are the ones people forget on a busy morning.

The API route, and why a clinic rarely needs it

OpenAI also offers a BAA for its API, the developer interface behind other companies' tools. OpenAI's help article describes a self-serve flow in the API Platform (Settings, then Organization, then General, then HIPAA compliance support) for organisations with an established history of API usage. The person accepting must be an organisation admin with authority to sign, HIPAA support cannot be switched off again in settings once enabled, and only endpoints OpenAI lists as eligible, with Modified Retention, may carry protected health information. That route suits a software vendor building a scribe or a portal, or a practice with a developer. For a clinic with no technical staff it is the wrong door: you would be taking on engineering, logging and access control that a packaged, covered product handles for you.

The useful point for a practice owner is on the other side of the transaction. When a vendor tells you their product is "built on OpenAI" and "HIPAA ready", ask for their own BAA with you, not a screenshot of OpenAI's. Your agreement is with the vendor you pay, and the vendor needs its own agreement behind it.

What a BAA does not cover

A business associate agreement is a contract about how the vendor handles protected health information on your behalf. It does not cover your own side of the arrangement. A few things remain yours even with a signed BAA:

• Staff access. Who has a login, who left last month and still has one, and whether accounts are shared.
• The minimum necessary rule. Even in a covered tool, paste only what the task needs.
• Accuracy. The vendor does not check the clinical content. A model can invent a medication dose or misread a result, and the clinician answers for what gets filed.
• Retention and deletion. Know how long prompts and outputs are kept and how to delete them.
• Your risk assessment. The tool belongs on your written risk analysis, with the date you reviewed it.

If you want the policy side drafted, the AI usage policy for a medical practice gives you wording to start from, including a section on which tools staff may use with patient data.

A five-minute check you can run today

Ask every person in the practice the same four questions and write down the answers.

1. Which AI tools have you used for work in the last month?
2. On which account and plan: personal, free, Plus, Business or something supplied by the practice?
3. Have you ever pasted anything that could identify a patient?
4. Have you used memory, connected apps or scheduled tasks?

The answers tell you where to start. Expect surprises: a hygienist using a personal Plus account to tidy notes, or a manager who believes the Business plan is covered, would not be unusual. Treat the findings as a baseline, not a disciplinary exercise. If someone did paste identifiable data into a plan without a BAA, record it, delete what you can, and ask your privacy lead or legal adviser whether it counts as an incident you need to act on.

A solo physician, a physio clinic and a multi-site group

A solo physician. Eligible for ChatGPT for Clinicians, willing to sign the BAA personally, and happy to follow the guide's feature exclusions. Patient letters can be drafted in the covered workspace; everything else stays out of it.

A physiotherapy clinic. Physiotherapists are not on the eligible clinician list OpenAI describes, so the clinic is left with Business for non-patient work and a different, covered tool for notes. See the AI scribe BAA and consent checklist for that side.

A multi-site medical group. This is the organisation ChatGPT for Healthcare is aimed at: central controls, many users, one agreement. OpenAI says pricing is based on ChatGPT Enterprise and depends on organisation size, so the group should expect a sales conversation and a configuration exercise, not a credit-card sign-up.

When to choose a different tool entirely

If your main use is patient notes, referral letters and inbox triage, a tool built for clinical documentation that signs a BAA as standard is often simpler than configuring a general assistant. The question is not "which chatbot is best?" but "which tool will my team use correctly every day, with a BAA in the file?" For costs, the tools a small practice pays for with a BAA compares options. If you would rather sort the jobs with someone else, that is what my AI implementation consultation covers.

A last point on honesty with patients. If AI touches their information in any way, your privacy notice and consent wording should reflect that, and a patient who objects should be able to opt out without losing care. Check that wording with your own adviser, as requirements differ by state and by service.

Follow-up questions from small practices

Is ChatGPT Plus HIPAA compliant?

No. Free, Go, Plus and Pro are not on OpenAI's list of HIPAA-eligible products, so no BAA covers them. Paying more raises usage limits, not protection. Without a BAA, protected health information must not be entered, whatever the privacy settings say.

Does turning off chat history make ChatGPT safe for patient data?

No. Training and history settings are privacy controls, not a HIPAA safeguard. HIPAA turns on whether a signed business associate agreement exists for the service you are using. Without one, the setting does not make patient details permissible.

Can I paste a patient note if I remove the name?

Only if the note is properly de-identified, which means removing all 18 identifier types HIPAA lists, not just the name. Dates, small locations, phone numbers and rare details can still identify someone. When in doubt, treat it as identifiable.

Does a signed BAA make my practice compliant?

No. A BAA covers the vendor's side. You still need policies, staff training, access controls, a risk assessment and a rule for which features may touch patient data. Treat the BAA as one required document in a longer set.

What should staff do if they pasted patient data by mistake?

Tell the privacy lead the same day, note what was pasted and where, delete the chat if the tool allows, and let the lead decide whether it counts as a reportable incident. A quick, blame-free report beats a hidden mistake.

Further reads

• AI Secretary for Doctors: Referral Letters, Inbox, Insurer Forms — Referral letters and inbox jobs a doctor can hand to AI.
• Is Claude HIPAA Compliant? Which Plans Offer a BAA — The same plan-by-plan check for Claude.
• HIPAA-Compliant AI Assistants for a Medical Office: BAA by Plan — A BAA check across several AI assistants for a medical office.
• ChatGPT for Chiropractors: What Is Safe Without a BAA? — What a chiropractor can safely ask ChatGPT without patient data.
• HIPAA-Compliant AI Receptionist: A BAA and Retention Checklist — BAA and retention checks for an AI phone or chat front desk.
• Best AI Receptionist for a Medical Practice: What Each Type Does — What each kind of AI receptionist does for a practice.
• Prior Authorisation With AI: Drafting Letters and Tracking Status — A back-office job where patient data and AI meet.
• ABA Session Notes With AI: From Raw Session Data to a Draft Note — A practice-owner workflow for turning trial data into a draft ABA note: de-identify or use a BAA tool, a prompt, a worked example and sign-off.
• ChatGPT for Medical Billing: Appeals and Claim Checks Without PHI — A redaction method, appeal and claim-check prompts, and the review step that lets a billing team use ChatGPT without pasting patient data.
• ChatGPT for Medical Notes: A Safe Drafting Workflow for Small Practices — Which note types a small practice can draft with ChatGPT, a de-identified workflow, and when only a BAA-covered scribe will do.
• HIPAA Compliance Policy for a Home Care Agency: Build It With ChatGPT — How to draft each HIPAA policy section with AI using no client data, then the exact items a lawyer and your compliance officer must check.
• Is Gemini HIPAA Compliant? What a Workspace BAA Covers — The Workspace BAA covers the Gemini app and Gemini in Workspace apps, not Gemini in Chrome. Here is the scope and the admin step that comes first.
• Medical Billing AR Follow-Up: Prioritise Aged Claims With AI — Turn an aged AR report into a ranked daily work queue, payer call scripts and consistent notes, using AI on de-identified data only.
• Is There an AI Tool for Pharmacists? What US and UK Teams Can Use — Yes: ChatGPT for Clinicians for verified US pharmacists, general business-plan tools elsewhere. What to use them for and what to keep out.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI help articles Getting a Business Associate Agreement for the OpenAI API, HIPAA eligible products and functionality, ChatGPT for Clinicians, ChatGPT for Healthcare and Health in ChatGPT; OpenAI business pricing page; HHS guidance on the Safe Harbor de-identification method (all read 4 November 2026).