Is ChatGPT HIPAA Compliant? What Health Tech Teams Need Before AI Touches PHI

Is ChatGPT HIPAA compliant? Not on its own. It can be used with PHI only on plans where OpenAI signs a Business Associate Agreement (BAA), such as ChatGPT for Healthcare, sales-managed ChatGPT Enterprise set up as a Regulated Workspace, and the API with Modified Retention. Free, Plus, Pro and Business plans carry no BAA. Even with a BAA, compliance depends on the controls your company runs around the tool.

Teams often pick the tool first and look for the BAA afterwards. Work in this order instead. Check what counts as PHI, decide what data the tool may see, confirm the vendor's BAA covers the exact plan and features you need, write the AI use policy, then turn the tool on.

If you handle PHI for providers or payers, you are a business associate yourself, so the same answers end up in your customers' security questionnaires.

Is ChatGPT HIPAA compliant?

No software product is HIPAA compliant by itself. HIPAA applies to organizations. The Privacy Rule limits how PHI is used and shared, and the Security Rule requires administrative, physical and technical safeguards. A tool can support those safeguards, and your company still has to run them.

Under HIPAA, covered entities and business associates need a BAA with any vendor that creates, receives, maintains or transmits PHI on their behalf, AI providers included. That vendor becomes a business associate too. When the company handing over PHI is itself a business associate, the agreement is a subcontractor BAA under the HHS business associate guidance.

A health tech company we work with chose the AI assistant it wanted for its team, then found the plan it was evaluating did not come with a BAA at the time. It ruled that option out before anyone put patient data into it and started checking another vendor's enterprise BAA.

It asked us to scope an AI use policy first. That policy sets what staff may and may not paste or attach, which tools are approved, and human review of the AI features inside its own product.

Which AI plans come with a BAA today?

As of 7 October 2026, OpenAI and Anthropic offer a BAA for ChatGPT for Healthcare, ChatGPT Enterprise and Edu, the OpenAI API, Claude Enterprise and the Claude API. Consumer plans, ChatGPT Business and Claude Team carry none. Both vendors also tie coverage to how the plan is configured, and eligibility changes often, so check their help pages again for your exact plan.

Is OpenAI's API or ChatGPT Enterprise covered?

An OpenAI BAA is available for most API services without an enterprise agreement. For ChatGPT, only sales-managed ChatGPT Enterprise or Edu accounts are eligible, and OpenAI does not provide a BAA for ChatGPT Business. Its list of HIPAA eligible products names the configurations that count, including ChatGPT for Enterprise with Regulated Workspace, ChatGPT for Healthcare and the API with Modified Retention. Individual clinicians on ChatGPT for Clinicians have a separate BAA flow.

PHI cannot go into ChatGPT Free, Plus, Pro or Business.

Is Claude HIPAA compliant?

Anthropic's support center says only Claude Enterprise plans can enable HIPAA, through a click-to-accept BAA. Team and individual plans (Free, Pro, Max) are excluded. Anthropic also offers a BAA for its API. A Claude API BAA signed before 2 December 2025 covers API use only and does not extend to the Enterprise plan.

Enabling HIPAA on Claude Enterprise does not bring every feature under the BAA. Some tools are covered only in specific configurations, so read the scope before you assume a feature is covered.

What a BAA does not cover

A BAA is a contract in which the AI vendor agrees to safeguard PHI as HIPAA requires. It covers the vendor. It says nothing about what your employees paste, whether anyone reviews outputs before they are used, or whether your retention settings match your own policies. Those are your controls.

A prompt sent to a model provider transmits whatever it contains, and a response that gets logged or cached is stored data. Send PHI only to vendors with a signed BAA that covers the plan and features in use, and only through configurations that match your written policy.

What your AI use policy needs to say

Your AI acceptable use policy should answer four questions.

• Which tools are approved? Name each tool and plan tier, and note which are covered by a BAA.
• What may staff paste or attach? Spell out what is allowed and what is not. If de-identified data is allowed, define what de-identified means in practice.
• Which outputs need human review? Any AI output that touches clinical content or patient records gets a defined review step before it is used.
• What is prohibited? Consumer plans, unapproved tools, and browser extensions that read clipboard or page content.

Under HIPAA's Safe Harbor method, data is de-identified only when all 18 identifiers are removed, including names, dates more specific than the year, IP addresses and URLs. A support ticket with the patient's name removed but the medical record number left in is still PHI.

Train the team on the policy, log each person's acknowledgment, and update it when a vendor changes its plans or features.

If you are working toward HIPAA compliance and have not written your AI use policy yet, our free HIPAA Compliance Checklist shows where it fits in the rest of your program.

What about the AI features in your own product?

The same BAA rule applies. If your product sends PHI to a model provider, that provider is your subcontractor and needs a BAA with you, whether the AI is an internal tool or a feature your customers use.

Buyers of AI in healthcare products also ask how the model is governed. ISO 42001, the management system standard for AI, is one way to show it.

Log what goes to the model and what comes back, and restrict access to those logs the same way you restrict the data itself. When an auditor or a customer asks how AI outputs are reviewed, those logs and your review records are the evidence.

How this shows up in your customers' security reviews

Enterprise customers' security reviews of health tech and AI vendors now include AI-specific questions like these:

• Is customer data used to train models?
• Which model providers see the data?
• Where is the data hosted, and how are tenants isolated?
• Is AI in scope of your SOC 2 report?

These questions arrive inside vendor risk assessments tied to a contract, so the answers need to exist before the questionnaire does. Anterior, a healthcare AI company, reached HITRUST in 7 weeks with us to meet a customer's security demands. ReadMe cut its security questionnaire responses from 6 business days to 2 with our GRC platform administration.

A short checklist before anyone pastes patient data

1. Find out what is in use now. Ask which AI tools and plans people already use for work, including personal accounts.
2. Map the data. Write down which types of PHI the tool might see, and from which systems.
3. Check the plan. Confirm the exact plan your company is on and whether the vendor's BAA covers it and the features you will use.
4. Sign the BAA. Execute it between your company and the AI vendor, covering the services in scope.
5. Write the AI use policy. Cover approved tools, allowed data, prohibited actions and human review.
6. Configure the controls. Set retention, access and audit logging to match the policy.
7. Train the team. Everyone who uses the tool reads and acknowledges the policy.
8. Update your questionnaire answers. Add the AI answers to your security review library before a customer asks.

If PHI has already gone into a plan with no BAA, treat it as a possible breach. Under 45 CFR 164.402, a disclosure HIPAA does not permit is presumed to be a breach unless a documented risk assessment shows a low probability that the PHI was compromised. A business associate with a breach of unsecured PHI must notify its covered entity customer without unreasonable delay, and no later than 60 calendar days after discovery, under 45 CFR 164.410. Check each customer BAA for its own notice deadline as well.

Need help with AI and HIPAA?

We work as the embedded security and compliance team, or the vCISO, for health tech companies. We scope the AI use policy, check that the BAA covers the plans and features you actually use, and get your security review answers ready before the next enterprise deal. Talk to an expert about your AI use policy.

Cancel anytime. If you're not saving 100+ hours, you don't pay.