Is ChatGPT HIPAA Compliant? Which OpenAI Products Come With a BAA (October 2026)

Is ChatGPT HIPAA compliant? Not by default, and no AI product is compliant on its own. OpenAI signs a business associate agreement only for ChatGPT for Healthcare, ChatGPT Enterprise with a Regulated Workspace, ChatGPT for Clinicians, its FedRAMP offerings and the API. ChatGPT Free, Go, Plus, Pro and Business come with no BAA, so protected health information should never go into them.

HIPAA does not certify software. It regulates covered entities, meaning health plans, clearinghouses and providers that conduct standard transactions such as claims electronically, and the business associates that handle protected health information (PHI) for them. The practical question is narrower: will OpenAI sign a business associate agreement (BAA) for the exact product you use, and have you configured and governed it the way that agreement assumes?

The answers below come from OpenAI’s HIPAA, BAA and product articles and from the HHS rules on business associates and de-identification, all checked on 14 October 2026. This is general information, not legal advice. For how each plan treats training, retention and residency more broadly, see our plan-by-plan look at ChatGPT and confidential data; for the deployment options open to a hospital, see the healthcare AI brief.

The verdict for each ChatGPT tier

OpenAI’s help center names six HIPAA-eligible offerings, each available with its Business Associate and Healthcare Addendum. Every other ChatGPT plan sits outside that list.

Every “yes” in that table carries two conditions. The agreement covers only the products and features OpenAI lists, and OpenAI warns that a feature’s availability does not by itself decide whether a given use is covered. And a BAA settles the vendor’s side only. The covered entity still runs its own HIPAA program around the tool.

What HIPAA requires before PHI reaches an AI vendor

The rule that decides most of this is 45 CFR 164.502(e). A covered entity may let a business associate create, receive, maintain or transmit PHI on its behalf only after obtaining satisfactory assurance that the associate will safeguard it, documented in a written contract or arrangement. Section 164.504(e) sets out what the contract must contain, including:

• the uses and disclosures of PHI the vendor may make, with nothing the covered entity could not do itself;
• appropriate safeguards, Security Rule compliance for electronic PHI, and reporting of breaches and other uses the contract does not allow;
• the same restrictions for any subcontractor that touches the PHI;
• access for HHS to the vendor’s records, and return or destruction of the PHI when the contract ends.

An AI service that receives prompts containing PHI is exactly this kind of vendor. HHS’s business associate guidance, last reviewed on 30 July 2026, lists a third-party AI chatbot on a provider’s patient portal among its examples. Its cloud computing guidance goes further: a cloud provider that stores only encrypted ePHI without holding the key is still a business associate, and a covered entity that uses one for ePHI without a BAA is in violation of the HIPAA Rules.

HHS also settles the certification question. Asked which cloud providers offer HIPAA-compliant services, its Office for Civil Rights replies that it does not endorse, certify or recommend specific technology or products. When any AI tool, ChatGPT included, is described as “HIPAA compliant”, that is the vendor describing itself.

Is ChatGPT Enterprise HIPAA compliant?

It can hold PHI, but only in a specific configuration. OpenAI’s eligible list names Enterprise with a Regulated Workspace rather than a standard Enterprise workspace, and its BAA page sends sales-managed Enterprise and Edu customers to its sales team for the agreement. OpenAI’s May 2026 feature sheet describes the Regulated Workspace as a preconfigured workspace in which features that do not support HIPAA compliance are switched off by default.

ChatGPT for Healthcare is the purpose-built version. OpenAI describes it as an enterprise edition for clinicians, administrators and researchers, priced on the ChatGPT Enterprise basis, with controls such as RBAC, SAML single sign-on, audit logs and customer-managed encryption keys, plus clinical search with citations. Its Epic plugin can bring authorized chart data into a conversation read-only, following existing Epic and workspace permissions.

What the agreement excludes matters as much as what it covers. As of 14 October 2026, OpenAI lists these as outside the BAA, to be used without PHI:

• improved memory, which is off by default in Healthcare and Regulated Workspaces;
• Sites, Codex tasks in the cloud and Codex Security administration;
• browser use and network access for Work in the cloud, and event-triggered scheduled tasks;
• internet access in Canvas and code blocks, and workspace agents shared with agent-owned connections.

OpenAI also says that features missing from its covered list are not automatically covered, and that newly released functionality may not be listed yet. Admins can open excluded features to RBAC groups, but only for work that involves no PHI, so review feature access whenever OpenAI ships something new. For what Enterprise seats cost next to a private deployment, see our ChatGPT Enterprise cost comparison.

Is ChatGPT Business HIPAA compliant?

No. OpenAI’s BAA page states it directly: it does not offer a BAA for ChatGPT Business. Business data is excluded from model training by default, according to OpenAI’s enterprise privacy page, but a training commitment is not the written assurance HIPAA requires, and no Business setting changes that.

A Business workspace is a company tool, but being one does not clear it for patient data. A practice that wants ChatGPT with PHI has three routes instead: ChatGPT for Healthcare through OpenAI’s sales team, individual ChatGPT for Clinicians accounts, or an application built on the API under its own BAA.

ChatGPT for Clinicians is free for verified US physicians, nurse practitioners, physician assistants and pharmacists, and each can sign a BAA in the product under Settings, Agreements. OpenAI adds a condition: share PHI only if a BAA is in place and you are authorized to sign one for the account. An employed clinician usually handles PHI on behalf of the employer, which is the covered entity, so check with its privacy office first. For several users or central administration, OpenAI points to ChatGPT for Healthcare.

Is the ChatGPT API HIPAA compliant?

The API is OpenAI’s developer platform, separate from the ChatGPT app, and the most accessible route to a BAA. OpenAI’s help center sets out the process and its limits:

1. No enterprise contract is needed. An organization admin accepts the standard Business Associate and Healthcare Addendum in the API Platform under Settings, Organization, General, HIPAA compliance support.
2. Eligibility is earned. Self-serve enrollment requires an established history of API usage; organizations not yet eligible can try later or contact OpenAI.
3. Retention must be configured. The organization must also be provisioned with Modified Retention, unless OpenAI specifies otherwise. Once HIPAA support is on, it cannot be switched off in the platform settings.
4. Only listed endpoints are covered. The list includes chat completions, responses, embeddings, audio transcription and speech, files, batches, fine-tuning jobs and the realtime API.

By default, OpenAI’s abuse-monitoring logs can contain prompts and responses for up to 30 days, and its zero data retention and modified abuse monitoring controls need OpenAI’s prior approval. With the BAA and the required retention setting in place, OpenAI says listed endpoints may process PHI even where data is retained.

The larger change is that the API makes you the owner of the application. OpenAI’s FAQ says accepting a BAA does not by itself make an application HIPAA compliant. Your code decides who sees which records, what goes into each prompt, what gets logged and for how long, and how a breach would be detected.

Plus, Pro, Free and the Health tab

None of Free, Go, Plus or Pro appears on OpenAI’s HIPAA-eligible list, and moving from a free plan to a paid one changes nothing about that. An employer also has no logs, no retention control and no contract covering a personal account.

Health in ChatGPT needs its own note because the name invites confusion. It is a tab for US adults on Free, Go, Plus and Pro, where people connect their own medical records and Apple Health data. OpenAI says connected records, and conversations that use them, are not used to train its foundation models or to target ads, and that data synced from a source is deleted within 30 days of disconnecting it, while conversation history stays until the user deletes it. Its FAQ also states that Health is not intended for clinical or covered-entity use and offers no BAA, and it directs covered entities and business associates to ChatGPT for Healthcare or ChatGPT for Clinicians.

That follows from how HIPAA is built. Its rules bind covered entities and their business associates, and a patient connecting their own records to a consumer app is neither, so the protection for that data comes from OpenAI’s Health Privacy Notice rather than from HIPAA. For anyone working for a provider or a health plan, the rule is simple: a patient’s information never goes into a personal ChatGPT account, whatever the plan.

What your organization still has to do

A signed BAA covers the vendor’s half. These duties stay with the covered entity or business associate, whichever product it chooses:

1. Risk analysis. 45 CFR 164.308(a)(1)(ii)(A) requires an accurate and thorough assessment of risks to ePHI. Add the AI workspace, its connectors and its data flows to that assessment before go-live.
2. Minimum necessary. Sections 164.502(b) and 164.514(d) require you to identify which workforce roles need which PHI and to limit access to match. For AI, decide who may use clinical connectors at all, and teach staff to include only the details a task needs.
3. Access control. Section 164.312(a) allows access to ePHI only for people and software that have been granted it, with a unique user ID for each person. Use single sign-on, map roles to RBAC groups and remove leavers promptly.
4. Audit controls. Section 164.312(b) requires mechanisms that record and examine activity in systems holding ePHI. On Enterprise and Healthcare workspaces, pull activity through the Compliance API into your security monitoring and review it regularly.
5. De-identification where you can. Under 164.514(b), data is de-identified only after an expert determination, or after removing 18 kinds of identifiers under the Safe Harbor method, among them names, all date elements except the year, geographic units smaller than a state and medical record numbers, with no actual knowledge that what remains could identify the patient.
6. Third parties inside the workspace. OpenAI warns against sharing PHI with third-party GPTs or connected apps unless the right permissions and agreements are in place, because they run under their own terms. Treat each one as a separate vendor decision; our AI vendor security questionnaire lists what to ask.
7. Training and rules for staff. Section 164.308(a)(5) requires a security awareness and training program. Name the one workspace approved for PHI, say plainly that personal accounts are never acceptable, and look for unapproved use; our shadow AI explainer covers how.

Keeping PHI on infrastructure you control

The other way to settle the HIPAA question is to take the AI vendor out of the PHI path. When an open-weight model runs on servers you operate, prompts, retrieved records and answers never reach an outside AI provider, so no AI vendor needs a BAA for that flow. Our guide for healthcare CIOs and CISOs compares the deployment patterns.

This does not take HIPAA away; it brings all of it in-house. Three points are easy to miss:

• The Security Rule now covers the AI stack. Risk analysis, access control and audit controls apply to the model servers, the vector index and the application, just as they apply to the EHR.
• Hosting and support can still create business associates. A cloud provider that runs your model servers is a business associate under HHS’s cloud guidance. HHS also lists IT vendors whose maintenance or support, in person or remote, requires access to ePHI. Plan support so that vendor staff never see PHI, or put the right agreement in place.
• Model quality and capacity become your job. You choose, test and update the models and size GPUs for peak demand, work a hosted service would otherwise absorb.

Administrative drafting, such as prior authorization letters and referral packets, is a sensible first workload because a person reviews every output before it is sent. Our notes on agents for healthcare administration cover where those workflows begin. If you also serve patients or sell devices in the EU, the EU AI Act guide for medical devices and hospitals explains the separate rules there.

How VDF AI fits

VDF AI is a self-hosted AI platform for running chat, private retrieval over your own documents and governed agents on your servers, in a private cloud or fully air-gapped. Prompts that contain PHI go to models you host, not to an AI vendor. VDF AI Router registers your on-premises model deployments, including Ollama endpoints, alongside any cloud models you choose to approve, and its air-gap mode keeps routing on local models only.

On-premises installations sign users in through Microsoft Entra ID, role-based access control comes with every plan, and audit logs record activity in storage you operate. Those are building blocks for the access and audit controls in 164.312, not a substitute for your risk analysis or policies. VDF AI holds no certifications, and HHS certifies no AI product, so whether a deployment meets HIPAA is decided by how you configure and run it. VDF AI Chat is the starting point for a ChatGPT-style assistant on your own hardware.