Last verified: October 2026, against Anthropic's help center and privacy center. General information, not legal advice — confirm with your compliance officer or attorney.
Is Claude HIPAA compliant? Claude can be used with patient information, but only on specific Anthropic products, after your organization accepts Anthropic's business associate agreement (BAA), and only in the features that agreement lists. The Free and Pro accounts many staff sign up for on their own are not among them.
Short answer (dated) from Anthropic's documentation
Short answer (verified October 2026): Anthropic's BAA article (opens in a new tab) says it provides a BAA for its "HIPAA-ready services": the first-party API, Claude in Microsoft Foundry and Claude Enterprise. Standard Enterprise plans "do not include BAA coverage without action from a Primary Owner," who must activate the HIPAA configuration and accept the BAA. The BAA covers only the organization that accepted it, and only the features Anthropic marks as covered. Free, Pro, Max and Team are not listed.
In practice: unless your practice runs a HIPAA-configured Enterprise organization or a BAA-covered API account, nobody should put protected health information (PHI) into Claude.
Free and Pro accounts vs Team, Enterprise and the API
Anthropic separates consumer plans (Free, Pro, Max) from commercial ones (Team, Enterprise, the API). The data rules differ, and so does the HIPAA answer.
| Plan | Training on your chats | Retention | BAA |
|---|---|---|---|
| Free, Pro, Max | Only if Model Improvement is on | Up to 5 years de-identified if on; 30 days if off | No |
| Team | Not by default | Kept until deleted; purged from back end within 30 days | No |
| Enterprise | Not by default | Same, plus custom retention | Yes, with the HIPAA configuration |
| API | Not by default | Inputs and outputs deleted within 30 days | Yes, once enabled by Anthropic |
The consumer rules come from Anthropic's privacy center: chats are used for training if you opt in, if they are flagged for safety review, or if you join a program such as Trusted Tester (training article (opens in a new tab)). With the setting on, Anthropic "may retain your data in a de-identified format for up to 5 years" (retention article (opens in a new tab)). The 30-day figure with the setting off is from Anthropic's data usage page (opens in a new tab). Incognito chats are not used to train Claude.
For commercial products, Anthropic says it will not use inputs or outputs by default (commercial training article (opens in a new tab)). The exception is feedback or bug reports someone chooses to send, which are kept for five years.
What Anthropic's BAA covers and its conditions
On a HIPAA-configured Enterprise organization, Anthropic lists these as covered: chat, projects and artifacts inside chat, file creation and code execution (without network access), voice, web search and Research. These are not covered:
- Skills. Available, but outside the BAA.
- Connectors, MCP servers and Enterprise Search. Third-party data flows are excluded.
- Claude in Chrome. Third-party data flows are excluded.
- Cowork in the cloud and Dispatch. Not covered or not available.
- Beta tools such as Claude Design, Claude Slides and Claude Docs, which aren't available yet for HIPAA-ready organizations.
For the API, the first-party Messages API is covered, but the Batch API, Files API, Skills API, code execution and web fetch are not. Anthropic's newest "Covered Models" also "require 30-day data retention" and aren't available with zero data retention (ZDR) unless Anthropic authorizes it (Covered Models retention (opens in a new tab)).
A BAA covers only the product named in it. A staff member's personal Pro account, a browser extension or a connector to an outside tool falls outside it, even if the practice also has Enterprise. Check what a BAA must include before you sign.
This is general information, not legal advice. Confirm how it applies to your practice with a qualified advisor.
Uploaded files, projects and data retention
Most HIPAA guides skip files, but this is where PHI slips in: a referral letter dragged into a chat, a spreadsheet of no-shows added to a project.
- Projects are covered inside chat on a HIPAA-configured Enterprise organization. Knowledge files added to a project inherit that coverage only there.
- The Files API is not covered and isn't accessible to HIPAA-ready API organizations.
- Retention: by default, Enterprise data "is retained indefinitely unless a custom retention period is set." A Primary Owner or Owner can set a custom period for chats and projects, with a minimum of 30 days (custom retention controls (opens in a new tab)).
- Deletion: deleting a chat removes it from history at once and from Anthropic's back end within 30 days (organization data retention (opens in a new tab)).
Set a retention period that matches your records policy instead of leaving the default.
Safe practice tasks without patient data
These are fine on any plan, because nothing in them identifies a patient:
- A plain-language rewrite of your consent form, kept for clinician review.
- Draft answers for your website FAQ, such as "Do you take walk-ins?"
- Five variations of a Google Ads headline for a new sports-physical service.
- A staff memo explaining a new check-in process.
Unsafe: pasting a referral letter, intake responses or a patient's portal message into a personal account. Removing the name doesn't fix it; read up on de-identified data first.
For a wider view, see using AI in practice marketing. Medfy drafts its SEO articles from MedlinePlus and PubMed, and the practice approves each one, so the clinical accuracy question has an owner. To see where your site stands, run a free marketing audit.
Rules for staff before anyone pastes PHI
- One approved account type. Practice-managed Enterprise seats with the HIPAA configuration, or no PHI at all.
- Personal accounts never touch patient work. Not for "just a quick summary."
- Keep uncovered features off. Skills, connectors and Claude in Chrome stay off for anyone who handles PHI.
- Minimum necessary. Paste only what the task needs, never a whole chart.
- Clinician review of anything clinical before it reaches a patient.
- Re-check Anthropic's coverage table every six months. It changes with new features.
The same tests apply across vendors. Compare ChatGPT and HIPAA and Gemini and HIPAA, or start from the HIPAA-compliant AI overview. For website forms and chat, Medfy keeps patient data in separate, encrypted regional databases under the practice's BAA; Medfy's patient-data safeguards explain how.
Frequently asked questions
Does Anthropic sign a BAA for Claude?
Yes, for its HIPAA-ready services: the first-party API, Claude in Microsoft Foundry and Claude Enterprise. On Enterprise, the Primary Owner must activate the HIPAA configuration and accept the BAA in the organization's settings. Free, Pro, Max and Team aren't on Anthropic's list as of October 2026, and only listed features are covered. Confirm the setup with your compliance officer or attorney.
Can staff use a personal Claude account for work?
Not for anything that involves patient information. Anthropic's BAA covers only the organization that accepted it, so a receptionist's personal Free or Pro account sits outside it. Even for non-PHI drafting, a practice-managed account is the better habit: you control who has access, and you can remove someone's access and data when they leave. Put the rule in your written AI policy.
Does Claude train on what I paste?
It depends on the plan. On Free, Pro and Max, chats are used for training only if the Model Improvement setting is on, a chat is flagged for safety review, or you send feedback; Incognito chats are never used. On Team, Enterprise and the API, Anthropic doesn't train on inputs or outputs by default, though feedback and bug reports you choose to send may be used.
Is Claude safe for drafting website content?
Generally yes, because a service page or FAQ shouldn't contain any patient's details. The risk is accuracy, not privacy. Check every clinical statement against a source such as MedlinePlus, PubMed or your specialty body, have a clinician approve it, and never paste in a patient's story, photo or message as raw material for a page.