What makes an AI tool HIPAA compliant?
Three things together. First, a contract: a covered entity may disclose protected health information (PHI) to a business associate only after obtaining satisfactory assurance that it will safeguard it, which in practice is a business associate agreement, or BAA (45 CFR 164.502(e), read 5 November 2026). Second, safeguards: the Security Rule requires the “confidentiality, integrity, and availability” of electronic PHI, through administrative, physical and technical controls (45 CFR 164.306, read 5 November 2026). Third, your own configuration and use. Compliance belongs to the deployment, not to the model.
Can ChatGPT, Claude or Gemini be HIPAA compliant?
Specific business products can, under a signed agreement. Here is what each vendor’s own page says:
| Vendor | Covered under a BAA | Named exclusions |
|---|---|---|
| OpenAI | API endpoints under a Business Associate and Healthcare Addendum | Web search with live internet access (OpenAI docs, read 5 November 2026) |
| Anthropic | Claude Enterprise once HIPAA is enabled, the first-party API, Claude in Microsoft Foundry, Claude Code in local mode with HIPAA settings | Console, beta features, Batch and Files APIs, web fetch (Anthropic, read 5 November 2026) |
| Workspace with Gemini in HIPAA Included Functionality; generative AI on Google Cloud’s agent platform | “Additional Google Services” (Workspace, updated 26 October 2026; Google Cloud, list updated 17 September 2025) |
The free app on a personal account is not covered by any of these agreements. Feature lists change often, so check the page on the day you sign.
Does HIPAA apply to every AI app that touches health data?
No, and that cuts both ways. HIPAA binds covered entities and their business associates. A legal analysis in the Journal of Law, Medicine and Ethics notes that when a patient types PHI into a chatbot, “the AI developer or vendor is neither a covered entity, nor a business associate”, so the data falls outside HIPAA (Rezaeikhonakdar, 2023, read 5 November 2026). The same tool becomes your compliance problem the moment your staff paste PHI into it, which is why shadow AI in a clinic is a breach risk, not a productivity quirk.
What should a buyer check before PHI goes into an AI tool?
Read the BAA against the exact features you plan to use, then confirm: retention and training terms; which features and connectors fall outside the agreement; access control and logging, ideally an agent audit trail for systems that act; where data is processed, a data residency question; and how staff are kept off personal accounts. Treat a “HIPAA certified” badge as marketing; the agreement and the controls are what count. Fold all of it into your AI vendor assessment, and where contracts are not enough, consider private AI deployment.