The HIPAA-Compliant ChatGPT Alternative: How to Choose an AI Tool Your Compliance Team Will Approve

A HIPAA-compliant ChatGPT alternative is an AI tool whose vendor signs a Business Associate Agreement (BAA) before your team uploads any protected health information (PHI), keeps that data out of model training, and runs on infrastructure your security team can check against named federal controls.

Most ChatGPT plans do not meet that standard. OpenAI's help center says only ChatGPT Enterprise or Edu customers with a sales-managed account are eligible for a BAA for ChatGPT, and that OpenAI does not offer a BAA for ChatGPT Business. If your team works with PHI, the plan it uses matters as much as the product name.

The question matters now because staff are already pasting work into AI tools. Whether "we can use ChatGPT" is a safe answer depends on which plan you buy, which features you turn on, and which contract stands behind them. This guide walks through what OpenAI's own pages say, what HIPAA requires of any AI vendor, the gaps a BAA leaves open, the security controls reviewers expect, a side-by-side comparison framework, the questions to ask vendors, and a step-by-step path to compliance approval.

Key takeaways

• "ChatGPT" is not a yes or no compliance answer. Personal plans and ChatGPT Business have no BAA, and OpenAI limits HIPAA eligibility to specific products, contracts, and features. Judge any ChatGPT alternative on those same terms.
• Any AI vendor that processes PHI for you is a business associate whether the data is encrypted or not. Get a BAA that meets 45 CFR 164.504(e), and confirm it flows down to the model provider and the cloud infrastructure.
• A BAA does not settle retention, training, or staff behavior. Ask for zero data retention by default and a written no-training commitment, give staff an approved tool, and do not rely on manual de-identification for real records.
• Map every vendor to the Security Rule's technical safeguards, ask for a named framework such as NIST 800-53 controls, and verify any FedRAMP statement down to the layer it describes.
• Compare every HIPAA-compliant ChatGPT alternative on the same rows (BAA, retention, feature scope, controls, infrastructure, model, usage terms, and document capacity), and then test your finalists on your own documents.
• Ten written questions cover the BAA, subcontractors, feature scope, retention, training, controls, infrastructure, audit evidence, incident reporting, and usage terms. A vague answer on any of them should stop the review.
• Bring compliance a full package of use cases, vendor evidence, a risk analysis, a reviewed BAA, a configuration plan, a one-page policy, and a pilot, and keep every record for 6 years.

Is ChatGPT HIPAA compliant? What OpenAI's own pages say

ChatGPT is not a single product. It is a family of plans, and each one comes with its own contract, retention rules, and training defaults. Whether a given plan can touch PHI depends on whether OpenAI will sign a BAA for that plan and on which features that BAA covers.

Which OpenAI products are eligible for a BAA?

OpenAI's page on HIPAA eligible products and functionality names six offerings that are available with a BAA:

• ChatGPT for Healthcare
• ChatGPT for Enterprise with Regulated Workspace
• ChatGPT FedRAMP
• ChatGPT for Clinicians
• The API with Modified Retention
• API FedRAMP with Modified Retention

Anything that does not appear on that list sits outside the BAA, no matter how the workspace is configured.

How do the ChatGPT plans compare on BAA, training, and retention?

The table below summarizes what OpenAI publishes for each plan type as of October 5, 2026. These pages change often, so confirm current terms with OpenAI before you rely on them.

Why are personal accounts the biggest exposure?

OpenAI's data controls article explains that personal plan users decide for themselves whether their conversations help train OpenAI models, using a setting called "Improve the model for everyone." That is a switch each user controls, not a control your organization manages. There is no BAA behind it either. A staff member who turns the setting off has changed how the data is used for training, but the account still has no contract that permits PHI.

Does ChatGPT Business cover HIPAA work?

No. Business workspaces receive enterprise-style privacy commitments, including no training on business data by default, according to OpenAI's enterprise privacy page. OpenAI also says plainly that it does not offer a BAA for Business. A workspace bought on a company card may feel like an organizational tool, but it does not make PHI use compliant.

Can an eligible plan still include features that are not covered?

Yes. OpenAI's HIPAA page lists functionality that is "not covered under the BAA," including improved memory and browser or network access for Work in the cloud, and it says PHI should not be entered when using those features. Buying an eligible plan is only the first step. Your configuration and your written policy both have to match what the BAA actually covers.

What does HIPAA require of any ChatGPT alternative?

HIPAA does not certify AI tools. HHS's cloud guidance says OCR does not endorse, certify, or recommend specific products, so no vendor can hand you a certificate that settles the question. Compliance comes from the contract you sign, the safeguards the vendor runs, and the way your team uses the tool.

Hathr.AI's overview of HIPAA-compliant AI tools sets out the minimum standard. The vendor signs a BAA, processes PHI only in an environment that BAA covers, and contractually excludes your PHI from model training and from human and machine review.

Why is an AI vendor a business associate?

HHS defines a business associate as a person or organization that creates, receives, maintains, or transmits PHI on behalf of a covered entity. Among its examples, HHS includes a third-party AI chatbot on a patient portal that handles PHI.

The HHS guidance on cloud computing closes the most common loophole. HHS says a cloud service provider that processes or stores ePHI is a business associate even if it holds only encrypted data and lacks the decryption key. The "conduit" exception, which covers transmission-only services such as the postal service, does not apply to a service with persistent access to PHI. An AI tool that reads your documents is processing PHI, so it falls on the business associate side of that line.

The same guidance adds that using a cloud service to maintain ePHI without a BAA violates the HIPAA Rules. An AI vendor is no different from any other cloud provider in that respect.

What must the BAA contain?

45 CFR 164.504(e) sets out the required elements, and HHS's sample business associate agreement provisions summarize them. Under those rules, the contract must do the following:

• Set the permitted and required uses and disclosures of PHI, and bar any others except as required by law.
• Require safeguards, including the Security Rule requirements for ePHI.
• Require the vendor to report unauthorized uses or disclosures, including breaches of unsecured PHI.
• Support patient rights to access, amendment, and an accounting of disclosures, and open the vendor's records to HHS.
• Require the return or destruction of PHI at termination where feasible, and allow you to terminate for a material violation.
• Flow the same restrictions down to any subcontractor that handles the PHI.

For an AI tool, the last point carries the most weight. The model provider and the cloud host behind the product are both subcontractors, and your PHI passes through each of them. Ask the vendor to confirm in writing that its BAA chain covers both.

What gaps does a BAA leave open?

A BAA allocates responsibility between you and the vendor. It does not tell you how long the vendor keeps your data, whether it trains on that data, or whether your staff actually use the approved tool. Four gaps come up in most compliance reviews, and each one needs its own answer.

How long does the vendor keep your data?

Every copy of PHI a vendor keeps is another copy that can be breached or mishandled. OpenAI's enterprise privacy page, for example, says API inputs and outputs may be kept for up to 30 days to provide the services and identify abuse, unless zero data retention is approved for an eligible endpoint.

The standard to ask for is zero data retention by default, which means prompts and documents are processed and then not stored afterward. Get that commitment in writing, and make sure it explains how retention applies to uploaded files and to system logs.

Is your data ever used to train models?

Ask for a written no-training commitment that also covers user feedback. Feedback is easy to overlook. OpenAI's data controls article notes that when a personal plan user rates a response, the entire conversation may be used for training, even after the user has opted out.

What happens when staff use AI tools you have not approved?

An approved tool only protects you if people actually use it. If staff keep personal AI accounts open in another browser tab, PHI will eventually end up there. The practical fix is to give them an approved option that handles their real work well, along with a short written policy that names it.

Can you de-identify PHI and keep using a consumer tool?

Some teams plan to strip out identifiers and keep using a consumer tool. HHS's de-identification guidance recognizes two methods for doing that:

• Safe Harbor removes a specified list of identifiers and requires that you have no actual knowledge that the remaining data could identify the person.
• Expert Determination requires a qualified expert's documented analysis of the data.

In practice, clinical notes and payer packs are full of names, dates, and record numbers, and scrubbing them by hand is easy to get wrong. For real records, use a tool that is covered by a BAA instead.

Which security controls will your compliance team want to see?

Once the contract is in place, your security team maps the vendor's safeguards to the HIPAA Security Rule and to a named control framework. Three areas usually come up in that review.

HIPAA technical safeguards

The Security Rule's technical safeguards at 45 CFR 164.312 include unique user identification, automatic logoff, encryption and decryption, and audit controls that "record and examine activity in information systems that contain or use electronic protected health information." Ask the vendor which of these it handles and which ones your team must configure yourselves.

HIPAA also requires you to retain Security Rule documentation for 6 years from the date it was created or the date it was last in effect, whichever is later. Add the vendor review, the signed BAA, and your AI use policy to that file.

NIST 800-53 controls

NIST SP 800-53 Revision 5 is the federal catalog of security and privacy controls for information systems and organizations. A vendor that follows NIST 800-53 controls gives your security team a shared vocabulary for access control, audit logging, and incident response, which makes the review faster and the answers easier to verify.

FedRAMP High Infrastructure

FedRAMP is the federal government's program for assessing cloud services. The FedRAMP Marketplace lets you look up a cloud service offering and see its impact level, such as Moderate or High.

When a vendor says its tool is built on "FedRAMP High Infrastructure," it means the tool runs on cloud infrastructure that was assessed at the FedRAMP High level. That statement describes the infrastructure layer of the product, so ask each vendor exactly which layer its FedRAMP statement refers to.

Try Hathr.AI free

HIPAA-Compliant, built on FedRAMP High Infrastructure. BAA provided on signup. Zero data retention by default. No usage limits, no seat minimums.

Start a free trial →

How do you compare HIPAA-compliant ChatGPT alternatives side by side?

Use the same criteria for every tool on your shortlist, including any ChatGPT plan you are considering. Fill in each row from what the vendor publishes, and ask for written confirmation wherever the published answer is unclear.

Which rows deserve the most weight?

Two rows usually decide the outcome. Document capacity determines whether staff can work from the full chart history or have to split files and risk losing details between the pieces. Usage and seat terms determine whether a department can keep using the tool during a denial spike, and whether a small practice can start at all.

How should you test your finalists?

Once each finalist has signed a BAA, run the same tasks through every tool using your own longest documents. Good test cases include a full chart summary and a denial letter matched against the payer's policy. Record the results side by side so your compliance and operations teams are looking at the same evidence.

What should you ask an AI vendor before your compliance review?

Send these ten questions to every vendor on your shortlist and ask for written answers. Each question includes the kind of answer that should move a vendor forward.

1. Will you sign a BAA before we upload any PHI, and does it cover the exact plan we will buy? A good answer is yes, on signup or before first use.
2. Does your BAA flow down to your model provider and cloud host? A good answer is yes, with the subcontractors named.
3. Which features does the BAA cover, and which does it exclude? A good answer is a written list, with excluded features turned off by default.
4. What is your default data retention? A good answer is zero data retention by default.
5. Is our data ever used to train models, including through user feedback? A good answer is no, stated in writing.
6. What security framework do your controls follow? A good answer names a framework, such as NIST 800-53 controls.
7. What infrastructure do you run on, and at what FedRAMP level? A good answer gives a level you can check in the FedRAMP Marketplace.
8. Can you share an independent third-party security audit report or equivalent evidence? A good answer is yes, under NDA if needed.
9. How will you report security incidents and breaches to us? A good answer is a timeline written into the BAA.
10. Are there usage limits or seat minimums? A good answer is that there are none.

What red flags should stop the review?

Watch for these warning signs in vendor answers and marketing:

• A claim that the product is "HIPAA-certified," since HHS does not certify products.
• A "HIPAA-ready" label with no BAA behind it.
• A BAA that covers only some of the services you plan to use.
• Accuracy percentages that are not backed by a published study.

How do you get your compliance team to approve an AI tool?

Approval goes faster when you bring compliance a complete package instead of a product link. These seven steps build that package in the order reviewers usually need it.

Step 1: Define the use cases. List the tasks the tool will handle, such as chart summaries or denial appeals, and note the minimum PHI each task needs.

Step 2: Collect vendor evidence. Send the ten questions above to each vendor and file every answer with the review.

Step 3: Run a risk analysis. HHS's cloud guidance reminds covered entities and business associates that they must analyze risks to all ePHI they create, receive, maintain, or transmit. Document the threats, the safeguards that address them, and the person who owns each one.

Step 4: Complete the legal review. Check the BAA against 45 CFR 164.504(e) and agree on breach-reporting timelines. HHS's sample provisions note that the parties may add a stricter timeframe than the rule requires.

Step 5: Configure the tool before launch. Turn on single sign-on and role-based access, disable any features that fall outside the BAA, and confirm that audit logging works.

Step 6: Train staff and publish a one-page policy. The policy should name the approved tool, ban personal AI accounts and any tool without a BAA for PHI, list the features that are off limits, require human review of every output, and explain how to report a mistake without blame.

Step 7: Pilot, then expand. Start with one team and one workflow, and widen access once the pilot holds up under real use.

Frequently asked questions

Is ChatGPT HIPAA compliant?

Not by default. OpenAI offers a BAA for ChatGPT only through specific products and contracts, such as sales-managed Enterprise or Edu accounts and its listed healthcare offerings. It does not offer one for ChatGPT Business or personal plans.

Can staff use a personal ChatGPT account if they turn off model training?

No. Turning off "Improve the model for everyone" changes how the data is used for training, not the contract. A personal account still has no BAA behind it.

Does a BAA make an AI tool HIPAA compliant?

A BAA is required, but it is not enough on its own. You also need the vendor's safeguards, a risk analysis, the right feature configuration, staff training, and documentation.

Is an AI vendor a business associate if it encrypts our data?

Yes. HHS says a cloud provider that processes or stores ePHI is a business associate even if the data is encrypted and the provider has no decryption key.

Can we de-identify PHI and use any AI tool?

Only if the data meets HIPAA's Safe Harbor or Expert Determination standard. Clinical notes and case files are hard to de-identify reliably by hand, so most teams use a tool covered by a BAA instead.

What does FedRAMP High Infrastructure mean for an AI tool?

It means the tool runs on cloud infrastructure that was assessed at the FedRAMP High level. The phrase describes the infrastructure layer, so ask each vendor exactly what its FedRAMP statement covers.

How fast can a small practice start with a HIPAA-compliant AI tool?

With a self-serve tool that provides a BAA on signup and has no seat minimums, a small team can start a pilot as soon as its internal review is complete.

Bottom line: which ChatGPT alternative will your compliance team approve?

The real question is not which AI tool is most popular but whether your compliance team will sign off on it. That decision depends on a BAA that covers the plan and features you will use, zero data retention by default, a written no-training commitment, named security controls, verifiable infrastructure, and a policy your staff will actually follow.

Hathr.AI is built for that review. It is HIPAA-Compliant, built on FedRAMP High Infrastructure, following NIST 800-53 controls, and Powered by a Government-Approved Claude. Teams get a BAA provided on signup, zero data retention by default, and no usage limits, no seat minimums. They can upload up to 100,000 pages per workspace, analyze 1,500+ page documents accurately, and read full case files, chart histories, and payer packs.

Hathr.AI is trusted by 3,000+ organizations, including the U.S. Dept of Health & Human Services. See how it compares as a ChatGPT alternative at hathr.ai/hipaa-chatgpt.

Talk to our team

Team of 10 or more? Book a 20-minute Enterprise Interview to walk through security review, BAA, and rollout.

Government-approved Claude, NIST 800-53 controls, FedRAMP High Infrastructure.

Request an interview →