HIPAA and AI: Is ChatGPT HIPAA Compliant, and Which AI Tools Can Handle PHI?

Consumer ChatGPT is not covered by a business associate agreement (BAA), so staff should not put protected health information (PHI) into it. Some business versions of ChatGPT and the OpenAI API can be covered by a BAA under specific conditions. The same rule applies to every AI tool, whether it is an LLM, a chatbot or an AI agent. A tool can handle PHI only when the vendor has signed a BAA that covers the exact product and plan you use, and your organization configures and uses it correctly.

Is ChatGPT HIPAA compliant? The short answer

No, not in its consumer form. ChatGPT Free and Plus are consumer products, and consumer AI products generally do not come with a BAA. If a front desk coordinator pastes a patient's name and insurance denial into a personal ChatGPT account to draft an appeal letter, that is a disclosure of PHI to a vendor with no BAA.

Some OpenAI business products can be covered. OpenAI may offer a BAA for certain business products and the API, under conditions that differ by product, so confirm current eligibility and any conditions with OpenAI directly. Being on a paid plan does not mean a BAA is in place. Someone at your organization has to request it, sign it and confirm which features it covers.

ChatGPT Health and ChatGPT for clinicians. OpenAI's health-focused offerings need to be checked by name, because a consumer health feature and a clinician-facing product can carry very different terms. A product that is marketed for health use is not automatically covered by a BAA with your practice.

The one-line rule: if there is no BAA for the exact product you use, do not put PHI in it.

For related topics, browse more guides on healthcare communication and HIPAA.

What makes any AI tool HIPAA compliant

There is no HIPAA certification. HHS does not certify software, AI models or vendors as HIPAA compliant. A "HIPAA certified AI" badge is marketing. It is not a government status. The useful question is: "Will you sign a BAA for this product, and what does it cover?"

An AI vendor that touches PHI is a business associate. When a vendor creates, receives, maintains or transmits PHI on behalf of a covered entity, it is acting as a business associate and needs a BAA. This applies to an LLM API that receives prompts containing patient details, an AI scribe that records visits and a chatbot that collects symptoms on your website.

A BAA is necessary but not sufficient. Even with a BAA in place, your organization is still responsible for:

• Security Rule safeguards: access controls, audit controls, transmission security
• The minimum necessary standard, which means sending the AI only the data the task actually needs
• Deciding who on staff may use the tool and on which accounts

A billing team using a covered AI tool to summarize every patient's full chart, when it only needs claim numbers, can still violate minimum necessary.

De-identified data is outside HIPAA, but "removing the name" is usually not de-identification. HIPAA recognizes two methods: Safe Harbor, which removes a defined list of identifiers, and Expert Determination, in which a qualified expert certifies that the re-identification risk is very small. Safe Harbor covers far more than names. It includes dates related to the individual (other than year), small geographic units, phone numbers, email addresses, medical record numbers and other identifiers. "58-year-old seen on March 3 at our Elm Street location for a rare condition" can identify a patient with no name at all.

Other laws stack on top. State health privacy and consumer privacy laws may add requirements, and GDPR applies if you process data of people in the EU. HIPAA compliance does not imply GDPR compliance.

Which AI tools and LLMs offer a BAA

The table below shows what to check for each major vendor. Every cell is a placeholder because vendor terms change often. Confirm each one against the vendor's current BAA and covered-services list before relying on it.

Vendor	Product or plan	BAA available	Business data used for training by default	Notes
OpenAI	API	Confirm with vendor	Check current terms	Ask about data retention and other conditions
OpenAI	ChatGPT Enterprise / Business / Edu	Confirm with vendor	Check current terms	Ask which plans are eligible and how to request
OpenAI	ChatGPT Free / Plus	Not expected for consumer plans; confirm	Check current terms	Consumer plans
Microsoft	Azure OpenAI Service	Confirm with vendor	Check current terms	Check Microsoft's BAA and Product Terms
Microsoft	Microsoft 365 Copilot	Confirm with vendor	Check current terms	Check which Copilot experiences are in scope
Google	Gemini in Google Workspace	Confirm with vendor	Check current terms	Check the Workspace covered-services list
Google	Vertex AI	Confirm with vendor	Check current terms	Check the Google Cloud covered-services list
Anthropic	Claude API	Confirm with vendor	Check current terms	Ask about conditions
Anthropic	Claude for Work (Team / Enterprise)	Confirm with vendor	Check current terms	Ask which plans are eligible
Amazon	Bedrock	Check the AWS HIPAA-eligible services list	Check current terms	Check coverage for each model you use

The general pattern. BAAs usually cover paid business or enterprise tiers and cloud APIs, not consumer apps. Within a platform, they often cover only the services listed in the BAA or covered-services list. A practice can have a signed BAA with a cloud provider and still be outside it if staff use a preview feature or a consumer app from the same company.

"Free HIPAA-compliant AI." Free consumer tiers generally come without a BAA. Treat this as a pattern, not a rule, and use the table to see what to check for each vendor. If an AI tool is free and the vendor will not sign a BAA, it is not an option for PHI.

SOC 2 and HIPAA for AI. Buyers often search for companies with both SOC 2 and HIPAA compliance. A SOC 2 Type II or ISO 27001 report is useful evidence that a vendor runs tested security controls. It does not replace a BAA, and it does not mean the vendor will accept PHI.

To review non-AI software the same way, check whether popular tools are HIPAA compliant.

HIPAA-compliant chatbots: patient-facing vs. internal

"HIPAA-compliant chatbot" covers two different things, and they are often different products.

	Patient-facing chatbot	Internal staff assistant
Who types into it	Patients and website visitors	Front desk, billing, managers
Typical uses	Intake questions, appointment requests, FAQs, routing	Drafting, summarizing, finding policies
PHI exposure	Collects PHI directly from patients	Depends on what staff paste or connect
Main risk	Unvetted chain of vendors receiving patient data	Staff using unapproved tools

Patient-facing chatbots need the whole chain covered. The chatbot vendor, its hosting provider and any LLM it calls all receive PHI. Your practice needs a BAA with the chatbot vendor, and that vendor needs BAAs with its own subcontractors. Website chat may also raise questions about tracking technologies on your site, so review those with your privacy officer.

Questions to ask a chatbot vendor:

• Which LLM does the chatbot use, and from which provider?
• Has that model provider signed a BAA with you, the vendor?
• Where are transcripts stored, and in which region?
• How long are transcripts kept, and can we set or shorten retention?
• Who at your company can read transcripts, and is that access logged?

This guide focuses on AI and HIPAA. For website chat, see HIPAA-compliant live chat for websites. For patient messaging tools, see patient communication and messaging software, and for video visits, see HIPAA-compliant telehealth platforms.

Conversational AI in healthcare. Common operational uses include answering hours and insurance FAQs, collecting appointment requests and routing messages to the right department. A chatbot that gives clinical advice or triages symptoms raises regulatory questions beyond HIPAA, including possible FDA oversight of software that supports clinical decisions. For the wider picture, see the benefits, risks and ethics of AI in healthcare and current AI in healthcare statistics.

AI agents and agentic AI: what changes under HIPAA

What an AI agent is. A chatbot answers a prompt. An agent takes actions: it reads a conversation, drafts a reply, updates a record in another system or calls another tool to finish a job. "Agentic AI in healthcare" usually means AI that does multi-step work instead of only answering questions.

Why agents raise the bar. An agent often has standing access to channels, files and connected systems, not just the text someone pastes in. That makes the minimum necessary standard and role-based access more important. An agent that helps the scheduling team should not be able to read HR channels or billing exports.

Controls to require from an agent platform:

• Per-role permissions that define what each agent can see and do, and who it takes instructions from
• The ability to see every active agent and revoke one immediately
• An audit trail of agent actions, showing what it read, what it did and who asked
• Clear documentation of which external models process the data, and where

BAA scope gets harder. With agents, check coverage at three levels, each separately:

1. The agent platform itself
2. The underlying model provider the agent sends data to
3. Every integration the agent can read from or write to

A gap at any level means PHI can leave covered territory.

"Which AI agent platform is HIPAA compliant?" None is compliant by label. A platform is usable for PHI only when it signs a BAA that explicitly covers agent processing and the models involved, and when the customer configures permissions, access and retention correctly.

AI vendor checklist for privacy and IT officers

Use these questions in your vendor review and record the answers in your risk analysis.

1. BAA scope. Will the vendor sign a BAA? Does it name the exact product, plan and features you will use, including AI features and agents?
2. Training on your data. Is customer data used to train models? Can that be excluded by contract, not just by a settings toggle a user could change?
3. Retention and deletion. How long are prompts, outputs, transcripts and uploaded files kept? Can you set retention? What happens to the data when you cancel?
4. Security controls. Encryption in transit and at rest, SSO, 2FA, role-based access, audit logs, and data residency options if you need them.
5. Subprocessors. Which model providers and cloud services touch the data? Are they covered by downstream BAAs?
6. Evidence. Current SOC 2 Type II or ISO 27001 reports, penetration test summaries and a named security contact.

For the rest of your program, use the HIPAA compliance checklist for clinics.

Safe AI use policy for clinic staff

Most AI risk in a practice comes from well-meaning staff using whatever tool is open in their browser. A short written policy prevents most of it.

Publish an approved-tools list. Name each AI tool, the account type staff must use (organization account, not personal), and what kind of data is allowed. For example: "Tool A, organization account: no PHI. Tool B, covered by our BAA: PHI allowed for billing appeals only."

Give front desk and admin staff one plain rule. No patient names, dates of birth, record numbers, phone numbers, addresses or other identifiers in any AI tool that is not on the approved list. When in doubt, leave it out.

Low-risk uses without PHI:

• Drafting staff announcements, such as a holiday hours change
• Rewriting a policy in plainer language
• Summarizing non-patient meeting notes from an operations huddle
• Creating training material and quiz questions
• Drafting job descriptions or onboarding checklists

For more ideas, see these practical uses of AI for clinic teams.

Put AI into your existing HIPAA program. Cover AI tools in your HIPAA training and in your Security Rule risk analysis, and record your decision about each tool. Check whether HHS has issued AI-specific guidance or rules that apply.

Offboarding. When staff leave, remove their access to AI tools the same way you remove email and chat access. Organization accounts make this possible. Personal accounts do not.

Where AI fits in internal team communication

Many practices first encounter AI inside their staff chat and work tools rather than as a standalone app. That makes the chat vendor's BAA and AI controls part of your HIPAA review, not a separate topic.

Chanty is internal team chat and work coordination for clinic and practice staff, used alongside clinical systems. It is not a patient messaging tool or an EHR. HIPAA with a signed BAA is available on the Business and Enterprise plans. It is not available on the Free plan. AI agents are available only on Enterprise. They come with per-role permissions that control what each agent can see and who it takes instructions from, the ability to revoke an agent, and a real-time audit trail of agent actions.

Do not assume BAA coverage extends to AI agents or connected external models. AI agents and the external models you connect are outside Chanty's standard BAA, so using agents with PHI requires a separate written agreement covering those models, in place before PHI is involved. Chanty conversations are not used to train outside models. Under shared responsibility, the customer chooses which models and agents to connect and what they can access.

For plan-level details, see Chanty security, HIPAA and BAA details, or read about HIPAA-compliant chat for staff.

FAQ

Is ChatGPT HIPAA compliant?

Consumer ChatGPT (Free and Plus) should not be used with PHI, because consumer AI products generally do not come with a BAA. Some OpenAI business products and the API may be eligible for a BAA under conditions, so confirm current eligibility with OpenAI. No product is HIPAA compliant by itself. Compliance depends on the BAA and on how you use the product.

Can I use ChatGPT with patient information if I remove the patient's name?

Usually not. Removing the name alone does not de-identify data under HIPAA. Safe Harbor requires removing a defined list of identifiers, including most dates, small geographic areas and record numbers, or you need an Expert Determination. If the data is not de-identified, it is still PHI.

Does OpenAI sign a BAA for ChatGPT Enterprise or ChatGPT Business?

OpenAI may offer a BAA for some business products, but eligibility and conditions can vary by plan and change over time. Ask OpenAI directly and get the signed BAA before any PHI is used.

Is there a free HIPAA-compliant AI tool?

Free consumer AI tiers generally come without a BAA, so they are not suitable for PHI. Any free tool you consider for PHI needs a signed BAA that covers it, which free tiers rarely offer.

Is a HIPAA-compliant chatbot the same as a HIPAA-compliant AI agent?

No. A chatbot answers questions. An agent takes actions across data and systems. Agents need tighter permissions, revocation, audit trails and BAA coverage that extends to the agent platform, the model provider and every connected integration.

Can therapists use AI tools for session notes under HIPAA?

Only with a tool whose vendor has signed a BAA covering that use, and with your practice's safeguards in place. Psychotherapy notes have additional protections under HIPAA. State mental health confidentiality laws and licensing board rules may also apply. For whether HIPAA covers your practice at all, see does HIPAA apply to therapists.

Do personal injury law firms need HIPAA-compliant AI when handling medical records?

It depends on the firm's role. A firm that receives records from its own clients is generally not a covered entity. A firm working on behalf of a covered entity may be a business associate. Professional confidentiality duties and state rules apply either way, so it is sensible to use AI tools with BAA-level protections for medical records and to check any guidance from your state bar.

Does SOC 2 certification mean an AI tool is HIPAA compliant?

No. A SOC 2 Type II report attests to security controls. It does not mean the vendor will sign a BAA or accept PHI. You need both the evidence and the BAA.

Can an AI tool be compliant with both GDPR and HIPAA?

A tool can support both, but they are separate laws with different requirements. HIPAA requires a BAA. GDPR requires a data processing agreement, a lawful basis and rules for transferring data out of the EU, among other things. Check each one separately.

Who is responsible if staff paste PHI into an unapproved AI tool?

The covered entity remains responsible for its workforce. The incident may require a breach risk assessment and notification, and the staff member may face sanctions under your sanctions policy. An approved-tools list, training and organization accounts make this much less likely.