Healthcare providers, practice managers, and compliance officers are asking a deceptively simple question: can we use ChatGPT? The answer matters, because clinicians and staff are already using AI to draft patient messages, summarize notes, and cut administrative time — and every one of those tasks can involve protected health information.

This guide answers the question thoroughly and practically. It explains what HIPAA actually requires, whether OpenAI will sign a Business Associate Agreement, the difference between consumer and enterprise AI products, how Microsoft Copilot and Claude compare, and the concrete steps a healthcare organization can take to use AI safely.

Executive Summary

ChatGPT is not automatically HIPAA compliant, and the free and personal versions should never be used with protected health information (PHI). No AI tool is “HIPAA compliant” on its own — compliance depends on the product tier you use, whether the vendor signs a Business Associate Agreement (BAA), how the tool is configured, and how your staff use it.

OpenAI does offer a BAA for its regulated healthcare offerings, including the OpenAI API and its purpose-built healthcare/enterprise products, but not for consumer ChatGPT (Free, Plus, or Team as typically sold). Microsoft covers Microsoft 365 Copilot under its existing Microsoft 365 BAA for eligible enterprise customers. Anthropic makes Claude available to HIPAA-regulated organizations through enterprise agreements and offers a BAA for eligible plans.

The bottom line: with the right enterprise plan, a signed BAA, correct configuration, staff training, and a governance policy, healthcare organizations can use AI with PHI. Without those, entering PHI into a consumer chatbot is a HIPAA violation waiting to happen. ITCS helps healthcare organizations adopt AI the safe way.

Key Terms

PHI (Protected Health Information)

Individually identifiable health information — anything that links a person to their health, care, or payment for care.

BAA (Business Associate Agreement)

A HIPAA-required contract between a covered entity (or business associate) and a vendor that handles PHI on its behalf, binding the vendor to HIPAA’s safeguards.

Covered entity

Health plans, healthcare clearinghouses, and most healthcare providers.

Business associate

A vendor that creates, receives, maintains, or transmits PHI for a covered entity.

Why Everyone Is Asking This Question

Generative AI arrived in healthcare workflows organically. A physician pastes a rambling note and asks for a concise summary. A front-desk employee drafts a patient reminder. A billing specialist asks a chatbot to explain a denial. Each use is genuinely helpful — and each can involve PHI.

The problem is that consumer AI tools were not built as healthcare systems. Depending on the product and settings, the data users type in may be retained, may be reviewed by staff to improve the service, or may inform future model training. That is fundamentally incompatible with HIPAA when the data is PHI and no BAA is in place.

So the real question is not “is ChatGPT good or bad?” It is “under what conditions, if any, can we use this tool with patient information without violating HIPAA?” That is a compliance and configuration question — and it has a clear answer.

Common Scenarios: Compliant or Not?

Abstract rules are easier to apply when you see them against everyday situations. The table below shows how common healthcare AI uses shake out. The pattern is consistent: the same task can be a violation or perfectly fine depending on the tool, the BAA, and whether real PHI is involved.

The through-line: strip out PHI, or use a BAA-covered and properly configured tool. Everything else is where violations happen.

What HIPAA Actually Requires

HIPAA’s Privacy Rule and Security Rule set the ground rules for using PHI with any technology, including AI.

The Privacy Rule limits how PHI may be used and disclosed and generally requires that you disclose only the minimum necessary information for a task. Sharing PHI with a vendor (like an AI provider) is a disclosure that must be permitted and, in most cases, governed by a BAA.

The Security Rule requires covered entities and business associates to protect electronic PHI through administrative, physical, and technical safeguards — including access controls, audit logging, encryption, and a documented risk analysis. Notably, HHS issued a Notice of Proposed Rulemaking in January 2025 that would significantly strengthen the Security Rule, with a final rule anticipated in late 2026. The proposal would make several currently “addressable” measures mandatory, including multi-factor authentication for access to ePHI systems, encryption of ePHI at rest and in transit, annual risk analysis, network segmentation, and regular vulnerability scanning and penetration testing. Healthcare organizations adopting AI should plan against this stricter baseline now.

The Business Associate framework is the piece most relevant to AI. If a vendor creates, receives, maintains, or transmits PHI on your behalf, that vendor is a business associate and you must have a BAA in place before sharing PHI. No BAA, no PHI — full stop.

Two additional points matter for AI specifically. First, HHS’s Office for Civil Rights (OCR) has emphasized, under Section 1557 nondiscrimination rules, that organizations using AI-driven decision-support tools must make reasonable efforts to identify and mitigate discrimination and keep a human in the loop. Second, “de-identified” data that meets HIPAA’s standard is no longer PHI and falls outside these restrictions — which points to one of the safest ways to use AI, discussed below.

Does OpenAI Sign a Business Associate Agreement (BAA)?

Yes — for the right products. OpenAI states that it can sign Business Associate Agreements to support customers’ HIPAA compliance, and it offers this for its API platform and its regulated healthcare/enterprise offerings, where customers can request a BAA through their account team. OpenAI also commits that business data from these enterprise products is not used to train its models by default.

Crucially, a BAA is not available for the ordinary consumer products — ChatGPT Free, ChatGPT Plus, and (as typically provisioned) ChatGPT Team are not covered for regulated PHI use. Availability and product names evolve, so confirm current BAA eligibility with OpenAI (or your IT partner) before relying on any specific plan.

The essential rule follows directly: you may only put PHI into an OpenAI product if you have executed a BAA that covers that specific product, and you have configured and used it accordingly. Without that, PHI must not be entered.

Can You Enter PHI into ChatGPT?

Here is the direct answer most readers came for.

Into consumer ChatGPT (Free, Plus, and standard Team): no. There is no BAA covering these products for regulated use, so entering PHI is a HIPAA violation risk regardless of how careful the individual is. This is the single most important takeaway to communicate to clinical and administrative staff.

Into an OpenAI product covered by a signed BAA (API-based applications or OpenAI’s regulated healthcare/enterprise offering): yes, with conditions. You must have the BAA executed, configure the environment appropriately (retention controls, access controls, audit logging), apply the minimum-necessary principle, keep a qualified human accountable for output, and train staff on correct use.

A practical and often overlooked point: a BAA covers the vendor’s handling of the data you send; it does not prevent staff from over-sharing PHI in the first place. Even with a BAA, sending an entire patient record when a de-identified summary would do violates the minimum-necessary principle and increases risk. Governance and training are not optional add-ons; they are part of compliance.

ChatGPT Enterprise vs. Consumer Versions

The gap between consumer and enterprise AI is the crux of the whole question. The table summarizes the practical differences for healthcare use. Confirm specifics with the vendor, since plans change.

The lesson is not “enterprise = automatically compliant.” It is that enterprise/regulated tiers make compliance possible by providing the BAA and the technical controls; your configuration, policies, and staff behavior determine whether you actually achieve it.

Microsoft Copilot and HIPAA

Many healthcare organizations already run Microsoft 365, which makes Copilot a natural question. Microsoft includes Microsoft 365 Copilot under its existing Microsoft 365 BAA for eligible enterprise customers — meaning organizations that already have Microsoft’s BAA and appropriate licensing generally have contractual coverage for M365 Copilot within their tenant.

Two caveats matter. First, coverage depends on the specific product and licensing: enterprise M365 Copilot is treated differently from the free/consumer Copilot at copilot.microsoft.com, which sits outside the BAA and should not receive PHI. Copilot Studio and other Copilot variants have their own considerations and connectors to verify. Second — and this echoes the OpenAI point — a BAA covers Microsoft’s processing, but it does not stop Copilot from surfacing PHI a user is not entitled to, or from a user pasting PHI into an unapproved surface. Because Copilot can retrieve information across your tenant, proper access controls, data governance, and sensitivity labeling inside Microsoft 365 are essential before you turn it loose on clinical data.

In short: M365 Copilot can be used with PHI for appropriately licensed enterprise customers under the Microsoft BAA, but only with correct tenant configuration, access governance, and staff rules.

Claude and HIPAA

Anthropic’s Claude is increasingly used in healthcare for its strong writing and reasoning. Anthropic makes Claude available to HIPAA-regulated organizations and offers a BAA for eligible enterprise/commercial plans (including its API and enterprise offerings), and it states that it does not use customer data to train its models under these terms. As with the others, consumer Claude plans are not the vehicle for regulated PHI use.

The same three-part rule applies to Claude as to every other tool: you need a signed BAA covering the specific product, correct configuration and access controls, and governed staff use with human oversight. Confirm current BAA availability and covered products with Anthropic or your IT partner before entering any PHI.

Safe Ways Healthcare Organizations Can Use AI

You do not have to choose between innovation and compliance. Healthcare organizations capture most of AI’s value while staying safe by following a few principles.

Prefer De-Identified or Synthetic Data

If information is de-identified to HIPAA’s standard, it is no longer PHI and can be used far more freely. For many tasks — drafting templates, learning to phrase patient education, summarizing de-identified cases — you never need real PHI.

Use Only BAA-Covered, Configured Tools for Any PHI

Restrict PHI to enterprise/regulated products with an executed BAA, correct retention and access settings, SSO, and audit logging.

Apply Minimum Necessary

Share only the PHI a task requires. Summaries and excerpts beat whole records.

Keep a Human Accountable

AI drafts; a qualified person reviews, verifies, and owns the output — especially for anything clinical or patient-facing. Verify every AI-stated fact, dosage, code, or citation.

Restrict High-Risk Uses

Do not let AI make final decisions about diagnosis, treatment, coverage, or eligibility without human review and required disclosures, consistent with OCR’s nondiscrimination expectations.

Train Staff and Govern with a Policy

Most incidents are accidental. Clear rules and training prevent the “quick paste into a chatbot” that causes breaches.

Log, Monitor, and Reassess

Maintain audit trails, monitor for sensitive data leaving approved tools, and include AI in your Security Rule risk analysis.

Building an AI Governance Policy for Healthcare

Safe AI use in healthcare rests on a written governance policy tailored to PHI. At minimum it should define which tools are approved for PHI (and which are prohibited), require a signed BAA before any PHI is shared with a vendor, enforce the minimum-necessary principle, mandate human review of AI output, set disclosure rules, and specify monitoring and enforcement. It should also fold AI into your existing HIPAA risk analysis and incident-response plan.

If your organization does not yet have such a policy, our companion guide walks through the components step by step and includes a free, editable template: AI Acceptable Use and Governance Policy for Businesses. Healthcare organizations should adapt that framework to add PHI-specific rules and BAA requirements.

Frequently Asked Questions

Is ChatGPT HIPAA compliant?

Not by default. Consumer ChatGPT (Free, Plus, standard Team) is not covered by a BAA and must not be used with PHI. OpenAI’s API and regulated healthcare/enterprise products can be used with PHI if you sign a BAA and configure and use them correctly.

Can I put patient information into ChatGPT?

Only into an OpenAI product covered by an executed BAA, configured with proper controls, and used under the minimum-necessary principle with human oversight. Never into the free or personal versions.

Does OpenAI sign a BAA?

Yes, for eligible products — the OpenAI API and its regulated healthcare/enterprise offerings. It is not available for consumer ChatGPT plans. Confirm current eligibility with OpenAI, as offerings change.

Is Microsoft Copilot HIPAA compliant?

Microsoft 365 Copilot is covered under Microsoft’s Microsoft 365 BAA for eligible enterprise customers, so it can be used with PHI when your tenant is properly licensed, configured, and governed. The consumer Copilot is not covered.

Is Claude HIPAA compliant?

Anthropic offers a BAA for eligible enterprise/commercial Claude plans and makes Claude available to HIPAA-regulated organizations. As with the others, you need the BAA, correct configuration, and governed use; consumer plans are not the path for PHI.

What happens if we use a non-compliant AI tool with PHI?

Entering PHI into a tool without a BAA can constitute a HIPAA violation and a reportable breach, exposing the organization to OCR penalties, breach-notification obligations, and reputational harm — even if the data is never publicly exposed.

Can we use AI safely without a BAA?

Yes — by using only de-identified or non-PHI data with those tools. De-identified information (meeting HIPAA’s standard) is not PHI and can be used far more flexibly.

Do we still need a policy if we have a BAA?

Absolutely. A BAA covers the vendor’s obligations; it does not govern how your staff use the tool. Governance, training, and monitoring are what prevent day-to-day violations.

How ITCS Helps Healthcare Organizations Deploy AI Safely

Adopting AI in a HIPAA environment is a compliance project as much as a technology one. ITCS helps South Florida healthcare organizations do it right: we assess where AI is already being used (including shadow AI), identify which tools and plans fit your workflows, help you execute the necessary BAAs, configure enterprise AI with the required security controls, fold AI into your HIPAA risk analysis and incident-response plan, and train your clinical and administrative staff so the rules hold up in daily practice.

Because AI safety is inseparable from your broader security posture, we integrate it with your Managed IT Services, Cybersecurity Services, and Compliance Services — and align everything with the strengthened HIPAA Security Rule requirements on the horizon.

Want to use AI without risking a HIPAA violation? Contact ITCS at 866-580-4826 to schedule a healthcare AI readiness assessment. If IT infrastructure and compliance are your first concern, start with our guide to HIPAA-Compliant IT Services for South Florida Healthcare Organizations.

Trust and sources: This article reflects HIPAA’s Privacy and Security Rules and guidance from the U.S. Department of Health and Human Services, Office for Civil Rights (OCR), including the January 2025 proposed Security Rule updates and OCR guidance on AI and nondiscrimination. Vendor BAA availability is based on each provider’s published enterprise/healthcare terms and can change; verify current terms directly with the vendor. This content is educational and not legal advice.