Can an AI tool be HIPAA certified?
No. HHS does not certify technology products, so HIPAA-compliant AI is something your organization puts in place itself. You need a signed business associate agreement (BAA) that covers the product tier and settings your staff use. Your risk analysis and staff rules must include the tool too.
The HHS guidance on HIPAA and cloud computing says: "OCR does not endorse, certify, or recommend specific technology or products." So if an AI tool calls itself HIPAA certified, that label is the vendor's own marketing.
Before anyone pastes protected health information (PHI) into a prompt, work through seven checks.
Why does an AI vendor need a business associate agreement?
In our reading of HHS cloud guidance, an AI vendor that processes PHI for you is a business associate. So sign the BAA before anyone shares PHI with the tool.
HHS treats a cloud service provider (CSP) that creates, receives, maintains or transmits electronic PHI (ePHI) for you as a business associate. That holds "even if the CSP processes or stores only encrypted ePHI and lacks an encryption key for the data." Both parties "must enter into a HIPAA-compliant business associate agreement (BAA)."
The HHS fact sheet on direct liability of business associates says business associates are directly liable for failing to comply with the Security Rule and for failing to enter into business associate agreements with their own subcontractors.
What are the 7 requirements for HIPAA-compliant AI?
HIPAA compliant AI comes down to seven requirements. The first six draw on HHS rules and provider documentation. The seventh is our recommendation.
- A signed BAA with the AI vendor
- A BAA that covers the exact product tier and features you use
- A risk analysis that includes the AI tool
- Only the minimum necessary PHI
- Access controls and audit controls
- Clear limits on what the vendor may do with your data
- A written staff policy for AI tools
1. Is there a signed BAA with the AI vendor?
Get a business associate agreement signed by your organization and the AI vendor before anyone enters PHI.
That agreement should also reach the vendor's own subcontractors. The HHS sample business associate agreement provisions require the business associate "to ensure that any subcontractors it may engage on its behalf" agree to the same restrictions.
2. Does the BAA cover the exact product and features you use?
A BAA covers only the product tiers and features it names, even when the vendor markets HIPAA compliant AI tools. Consumer plans fall outside the AI vendor BAAs we reviewed.
OpenAI's list of HIPAA-eligible products (October 2026) names specific healthcare, enterprise and API offerings. Some features are excluded even inside those products, and the list says "Features not listed here are not automatically covered under the BAA." Consumer ChatGPT is not on the list. For API use, OpenAI requires an executed BAA and Modified Retention. Our guide to whether ChatGPT is HIPAA compliant walks through each OpenAI plan.
Anthropic's BAA page for commercial customers (October 2026) covers its API and HIPAA-ready Enterprise plans. Your organization has to switch on HIPAA settings. And the BAA covers only the organization that accepted it. Consumer plans and a list of features are excluded, and data sent to third parties through connectors "isn't covered under Anthropic's BAA."
So a product can be covered on paper while staff use a personal account or an excluded feature. Picture a front desk lead who starts a patient letter at work, then finishes it at home on a personal account. No BAA covers that second half. These lists also change, so check the current version before each rollout.
3. Is the AI tool part of your HIPAA risk analysis?
Add the AI tool to your risk analysis when you adopt it. The Security Rule, at 45 CFR 164.308(a)(1)(ii)(A), requires an "accurate and thorough assessment of the potential risks and vulnerabilities to the confidentiality, integrity, and availability of electronic protected health information" held by your organization. HHS notes that "the Security Rule does not prescribe a specific risk analysis methodology," so extend the process you already use.
The HHS guidance on risk analysis says "The risk analysis process should be ongoing" and that an integrated process "is performed as new technologies and business operations are planned." HHS cloud guidance adds that "the type of cloud configuration to be used may affect the risk analysis and risk management plans of all parties." For an AI tool, record the tier and settings your team uses.
OCR actively enforces this requirement. In a case that did not involve AI, the HHS announcement of the Health Fitness Corporation settlement (March 21, 2025) reported a $227,816 settlement and called it the fifth enforcement action in OCR's Risk Analysis Initiative. OCR Acting Director Anthony Archeval called an accurate and thorough risk analysis "not only required" and "the first step."
4. Are you sending the AI only the minimum necessary PHI?
Send the AI only the PHI the task requires, and remove identifiers when the task does not need them. The Privacy Rule "generally requires covered entities to take reasonable steps to limit the use or disclosure of, and requests for, protected health information to the minimum necessary to accomplish the intended purpose."
The HHS minimum necessary guidance lists exceptions, including disclosures to a health care provider for treatment and uses made under a patient's authorization. Our reasoning is that drafting appeal letters and summarizing claim notes are not treatment disclosures, so the standard applies.
Picture a billing coordinator who pastes a whole patient chart into a chat window to draft one appeal letter. The appeal rarely needs a Social Security number, and most of that chart is PHI the task never required.
When identifiers are not needed, de-identify. HHS guidance on de-identification methods describes two methods under 45 CFR 164.514(b): Expert Determination, "a formal determination by a qualified expert," and Safe Harbor, the removal of 18 specified identifier types such as names and medical record numbers. HHS states that "the Privacy Rule does not restrict the use or disclosure of de-identified health information."
5. Are access and audit controls in place around the AI tool?
Only authorized staff should be able to reach the AI tool and its outputs, and you should be able to see who did what. The HHS summary of the Security Rule says systems must restrict access "to allow only authorized persons to access ePHI" and requires mechanisms to "record and examine activity in information systems that contain or use ePHI."
We recommend role-based access to any workspace that handles PHI, plus logs that someone reviews. Treat a summary the AI writes from PHI like the source record. Our reading is that the integrity standard applies when AI output is written back into a record: "Electronic measures must be put in place to confirm that ePHI has not been improperly altered or destroyed."
HHS calls the Security Rule "flexible, scalable, and technology neutral," and it does not dictate specific measures, so you choose what meets each standard. We recommend encryption. Under HHS cloud guidance, it does not change a vendor's status as a business associate. The proposed Security Rule would require encryption of ePHI at rest and in transit, with limited exceptions.
6. Are there clear limits on what the vendor can do with your data?
The BAA must spell out what the vendor may do with your PHI. HHS sample provisions say a business associate contract must "establish the permitted and required uses and disclosures of protected health information by the business associate" and provide that it "will not use or further disclose the information other than as permitted or required by the contract or as required by law."
For an AI tool, check four points in the BAA and the product settings:
- Permitted uses: can prompts or outputs be used to train or improve the vendor's models, or for anything beyond serving your requests?
- Retention: how long are prompts and outputs kept?
- Subcontractors: who else processes the data, under what terms?
- End of contract: must the vendor return or destroy PHI at termination, if feasible, as HHS provisions require?
7. Do your staff have a written policy for AI and PHI?
We recommend a short written policy that names the AI tools approved for PHI and says how staff may use them. HHS's Security Rule summary says a regulated entity must train all workforce members on its security policies and procedures.
The policy should ban PHI in personal or consumer accounts and name the features staff may not use. Once the AI policy sits inside your security policies, the training requirement covers it too. Review it whenever a vendor changes its covered list.
When is a BAA-covered cloud AI product enough?
In our view, a BAA-covered cloud product is often enough for drafting and summarizing once the covered tier and settings are in place. Keeping PHI in your own environment makes more sense when workflows touch many records or connect to core systems.
| Factor | BAA-covered cloud AI product | AI in your own environment |
|---|---|---|
| Who holds the PHI | The vendor processes it in its cloud under the BAA | It stays within systems your organization controls |
| BAA needed | Yes, with the AI vendor, covering the exact product and features | Yes, with any outside party that handles PHI for you, such as a hosting or support provider |
| Setup effort | Lower: sign the BAA and configure the covered settings | Higher: you build the integration and maintain it |
| Control over retention | Set by the vendor's terms and available settings | Set by your own policies |
| Feature limits | Excluded and beta features may be off limits | Limited to what you build and approve |
| Best fit | Individual drafting and summarizing by trained staff | High-volume workflows tied to your EHR or billing systems |
The deciding factor is how far PHI travels. If an automation pulls from your EHR all day, keeping it inside your environment means fewer parties and settings you depend on. For multi-step processes, see our intelligent workflow automation service.
How could the proposed Security Rule changes affect AI tools?
A proposed Security Rule update would raise the baseline for systems that hold ePHI, and we expect that to include AI tools that process PHI. It is not final.
OCR issued the proposed rule on December 27, 2024, and the Federal Register notice of the proposed Security Rule appeared January 6, 2025 (90 FR 898). According to the HHS fact sheet on the proposed Security Rule, proposals include encryption of ePHI at rest and in transit and multi-factor authentication, each with limited exceptions, plus a technology asset inventory and network map. Business associates would have to verify their technical safeguards to covered entities at least once every 12 months.
HIPAA Journal reported in July 2026 that the federal regulatory agenda now targets July 2027 for a final rule, though that date may move. List any AI tool you add in your technology asset inventory now.
How should a healthcare organization start using AI with PHI?
Start with one well-defined use case and run it through the seven requirements before you expand.
Then name one person who owns the AI policy and approves tools before staff use them. If nobody in-house has the time, a fractional Chief AI Officer can fill that role without a full-time hire. For assistants built into daily work, see our AI agents and copilots service.
If you want PHI to stay in-house, we build custom workflows to HIPAA requirements directly in your own environment, without your information ever touching OpenAI, Google, or Anthropic. Our automation work for medical practices covers intake, prior authorizations and claims follow-up, connected to your EHR and billing systems.
This article is general information and should not be read as legal advice. Have counsel review any BAA before you sign it.
Frequently Asked Questions
Is AI HIPAA compliant?
No AI tool is HIPAA compliant by itself, and OCR does not certify technology products. HHS treats a cloud provider handling ePHI for you as a business associate that needs a BAA, and we read AI vendors the same way. So HIPAA compliant AI depends on that BAA and on the safeguards your organization puts around the tool.
Is ChatGPT HIPAA compliant?
Only certain ChatGPT offerings are covered, and consumer ChatGPT is not one of them. As of October 2026, OpenAI lists specific healthcare, enterprise and API offerings as HIPAA eligible with a signed BAA. It also excludes some features even inside those products. Anything not on the list is not automatically covered, so check the current version before anyone enters PHI.
Is Claude HIPAA compliant?
Some Claude services are covered by Anthropic's commercial BAA. As of October 2026, it covers HIPAA-ready services such as the first-party API, Claude in Microsoft Foundry and Enterprise plans. Enterprise organizations must activate HIPAA settings, and the BAA covers only the organization that accepted it. Consumer plans and some features, including beta features, are excluded.
What happens if staff put PHI into a consumer AI tool?
That use has no BAA behind it, at least for OpenAI and Anthropic consumer plans. As of October 2026, consumer ChatGPT is not on OpenAI's HIPAA-eligible list, and Anthropic's commercial BAA excludes Claude Free, Pro and Max. The vendor has not agreed to the use limits and safeguards HHS sample provisions require. Stop that use right away and have your privacy officer document what happened.
Can staff put de-identified data into an AI tool?
Yes, if the data is properly de-identified. The Privacy Rule does not restrict the use or disclosure of de-identified health information. HHS recognizes Expert Determination by a qualified expert and Safe Harbor, which removes 18 types of identifiers such as names and medical record numbers. Data that only looks anonymous may still contain identifiers, so use one of these methods first.
Is the new HIPAA Security Rule in effect?
No. As of October 2026, the updated Security Rule is still a proposed rule, published in the Federal Register on January 6, 2025. It would require encryption of ePHI and multi-factor authentication, with limited exceptions. HIPAA Journal reported in July 2026 that the federal regulatory agenda targets July 2027 for a final rule, but that date is not certain.
Final Takeaways
- HHS does not certify AI tools, so HIPAA compliant AI depends on your BAA and your own safeguards.
- Sign a BAA that covers the exact product tier and features your staff use, and recheck the covered list before each rollout.
- Add every AI tool to your risk analysis when you adopt it, and apply minimum necessary and access controls to prompts and outputs.
- Put approved tools in a written staff policy and train your team on it.
- Add AI tools to your technology asset inventory now, and treat the Security Rule update as a proposal until a final rule is published.
If you are deciding which AI tools your staff can use with patient data, book a strategy call with Aprendio to talk through your first use case.
