Is ChatGPT "HIPAA compliant"? The short answer
No office should assume that ChatGPT in its standard form is safe for patient information. Whether any AI tool fits HIPAA depends on the contract, the settings and how your staff use it. A product name alone does not settle it.
HIPAA is the federal law that sets privacy and security rules for health information. This page covers what the rules say. It is not a verdict on any product. Other privacy laws, such as state laws, may also apply to your office, so ask a qualified professional about those too.
The rules protect what HHS calls protected health information, or PHI. That means individually identifiable health information that a covered entity or its business associate creates, receives, maintains or sends. In plain words, it is any detail about a patient's health, care or payment that points to a specific person. When PHI is stored or sent in digital form, HHS calls it electronic protected health information, or ePHI.
The rules apply to covered entities. HHS defines a covered entity as a health plan, a health care clearinghouse, or a health care provider that conducts certain billing and payment transactions electronically. A clearinghouse is a go-between that processes health billing data. Most healthcare organizations that bill electronically fall into this group.
OpenAI is the company that makes ChatGPT. Its terms change over time, so check OpenAI's current terms yourself before you decide anything. The same goes for every other AI vendor.
Why standard ChatGPT is a problem for PHI
The problem is not that the tool is bad. The problem is who ends up holding the information. When staff paste PHI into an outside service, that company may become a business associate under HIPAA. The law then requires a signed business associate agreement. Without one, pasting PHI into the tool is a risk to your office.
Think of everyday cases. A visit note. A referral letter. A billing question that includes a patient's name and insurance number. Each one holds identifiers. An identifier is a detail that points to a person, such as a name, address or date of birth. Using ChatGPT for any of these tasks puts that health data in front of an outside company.
Some plain definitions help here:
- An AI chatbot is a program you type questions into and get written answers from.
- Generative AI is software that writes new text based on patterns it learned.
- Large language models are the programs behind most chatbots. They are trained on huge amounts of text to predict and write language.
- A prompt is the text you type in.
- Data retention is how long a company keeps what you type.
Where a prompt goes, and how long the vendor keeps it, are questions for the vendor's current privacy policies and terms. Read them. Do not rely on memory, a blog post or a sales call.
The HHS page on the rule for electronic PHI makes clear that the protections cover digital records. Its Security Rule page says the rule sets national standards for electronic protected health information held by a covered entity or its business associate. A chat window that receives a patient's details is a place where ePHI can land.
So the safe question is not "is paid ChatGPT better than free ChatGPT?" The question is whether the vendor will sign a business associate agreement with you, and what that agreement covers. Ask. Read the answer in writing.
Business associates and the BAA, in plain words
A business associate is a person or company, outside your own staff, that handles PHI for you. HHS describes it as an entity that performs functions or provides services to a covered entity that involve creating, receiving, maintaining or sending PHI. Subcontractors count too. A company that handles PHI for one of your business associates is also a business associate.
The contract that ties this together is the business associate agreement, or BAA. It is a written contract. It is not a feature you switch on in a settings menu. Many offices mix this up.
HHS gave guidance on cloud services that shows how wide this reaches. In its cloud computing guidance, HHS says a cloud provider that stores or handles electronic PHI for a covered entity is a business associate. That is true even if the provider only holds encrypted data and has no key to read it. Having no key does not remove the duty to have a BAA.
Here is why that matters for AI. Online AI services that receive PHI from your office are, in many setups, run by an outside company. If the company handles PHI for you, the cloud guidance points to a business associate relationship. The details depend on the vendor and your facts, so ask a qualified professional about your own case.
HHS also says both sides carry duties. The covered entity and the provider must sign the agreement. The provider is bound by the terms of the contract and is directly responsible under the HIPAA rules. Your office stays responsible for its own side.
Two more plain terms:
- The Privacy Rule is the rule on how health information may be used and shared.
- Security safeguards are the steps that protect electronic PHI. The safeguards section below covers them.
A signed BAA is one piece of HIPAA compliance. It does not cover everything else your office has to do. That includes staff habits, logins and written policies. Many offices hear "we signed a BAA" and stop there. Do not stop there.
Plenty of offices hold such contracts already: with the billing company, the cloud storage vendor, the shredding service. Adding an AI vendor means adding one more, and checking how it fits in with the others. Keep a simple list of every business associate and what each one touches.
Paid plans, "HIPAA compliant AI" tools and what they cost
People ask whether a tool exists that meets HIPAA rules, how to get one and what it costs. They also ask about "ChatGPT for healthcare" and about enterprise plans, which are the higher business tiers some vendors sell. Some vendors sell healthcare AI products. Some say they will sign BAAs. This page does not name plans, state prices or say what any product offers. Vendors change their terms, and a price quoted here would go stale. Check each vendor's current terms for what its enterprise plans include.
Ask each vendor these questions, in writing:
- Will you sign a BAA with our office?
- What does the BAA cover, and what does it leave out?
- What are your retention rules for prompts and files?
- Where is our data stored? This is called data residency, and it means the country or region where your files sit.
You can add more. Ask whether the vendor keeps an audit record of who used the tool and when. That tells you whether anyone can see what happened if something goes wrong. Ask whether the tool supports SSO, short for single sign-on, which lets staff use one office login for many tools. Ask whether it offers role-based access, which means each person sees only what their job needs.
Also ask about training on your data. This means whether your prompts are used to train the AI model, which is the trained program that writes the answers. Read the answer in the vendor's current terms, not in an ad.
A signed agreement does not make a whole office meet HIPAA rules. Training, access, device settings and written policies still fall to you. And "safe" is not a promise anyone can make about an AI tool. Be cautious of any page, including this one, that tells you otherwise.
The rules also keep moving. HHS lists a proposed Security Rule update dated January 6, 2025. Check its current status on the HHS site before you build a plan around it.
What your office can do to use AI more safely
These steps hold for any AI, whether it runs online or on your own server.
1. Do a risk analysis
A risk analysis is a written review of where PHI lives in your office and what could go wrong. HHS and the Office of the National Coordinator for Health IT offer a Security Risk Assessment Tool, listed on the HHS page. HHS says it is useful for small and medium-sized practices. Add any AI tool to the review before staff start using it. The use of AI for clinical documentation, such as drafting visit notes, deserves its own line in that review, because those notes sit in the patient's health record.
2. Set safeguards
The rule calls for administrative, physical and technical safeguards. Administrative means policies and training. Physical means who can get to the computer or server. Technical means things like logins and encryption, which scrambles data so outsiders cannot read it.
3. Control who can reach what
Access controls decide who can open which files and tools. Give each person their own login. Give people only the access their job needs. Remove access the day someone leaves.
4. Write a short rule on prompts
Staff often ask about "safe ChatGPT prompts." Be careful here. Removing a patient's name does not always remove every detail that points to a person.
HIPAA describes a method called Safe Harbor for making data de-identified, meaning it no longer points to a person. It involves removing 18 HIPAA identifiers, such as names and dates tied to the patient. Free-text notes are hard to scrub by hand. A date, a rare condition or a small town can still point to one person. So a "safe prompt" is not a guarantee. Ask a qualified professional before you treat any text as de-identified.
Put a short list in writing: what may never be typed in, and who to ask when unsure. Keep it to one page. Tasks that fall outside that list, such as drafting a general letter with no patient details, can have a looser rule.
5. Train staff and check real use
Walk through the rules with staff. Then check how people actually use the tool. Map each step of your AI workflow before you pick a tool, so you buy for the work you really do. Review it every so often, because habits drift.
Putting AI inside a process is a different job from giving staff a chat screen. If you want it wired into a workflow, see how process automation works.
Private AI on your own server: one way to keep PHI in-house
One option is to run the AI on a server your office owns. Wirestack sets up private AI this way. Staff get a screen like ChatGPT and can ask questions about the office's own files. Prompts and documents stay on the office's network. The server goes online only to download models and updates. You can read the details on the private AI page.
Here is how this connects to the HHS guidance. Under that guidance, a cloud provider that handles PHI for you is a business associate. If the AI runs on your own server, there is no outside AI company in the picture.
That does not finish the job. The server is still a system that holds patient data. The same duties apply: access controls, logs of who used it, encryption, backups and written policies. Wirestack sets up logins and keeps the server inside your network. Wirestack does not certify any office as meeting HIPAA rules. Your office still owns its policies and its written review.
For home health and hospice offices, visit notes, referrals and care plans are full of patient details. Wirestack also builds software for home health agencies. See its home health work.
There are honest limits:
- Open models that run on an office server are smaller than the best cloud models. They do well at drafting, summarizing and answering from your files. For the hardest reasoning, the largest cloud models are still ahead.
- You buy the machine. A workstation costs thousands of dollars, and a server with several graphics cards costs more.
- Business ChatGPT accounts can be enough when nothing sensitive goes in. If staff only draft general letters with no patient details, a server may be more than you need.
Private AI suits offices where staff handle patient files every day and do not want another outside company holding that data. If that is not you, you may not need it.
Questions to ask before any AI tool touches patient data
Print this list and use it with every vendor and every internal pilot.
- Will this vendor sign a BAA, and what does it cover?
- Where do prompts and files go, and how long are they kept?
- Who in the office can use it, and how do we log them in?
- What is staff allowed and not allowed to type in?
- Who checks the answers, since AI can be wrong?
- Have we updated our written risk review and written policies?
If a vendor cannot answer in writing, treat that as an answer.
If you want to talk through your own office, get in touch. Wirestack gives a written plan before you spend anything.
This is general information, not legal, billing or compliance advice.
