Is Claude HIPAA Compliant? Which Plans Offer a BAA

Claude is not HIPAA compliant by default, and only one plan family can sign a BAA. Anthropic offers a HIPAA-ready Claude Enterprise, self-serve or sales-assisted, which the organisation's Primary Owner enables and accepts the BAA for, plus a BAA for its API. Free, Pro, Max, Ultra and Team cannot enable HIPAA, so patient data stays out.

The trap is the word "plan". A practice that pays $100 or $200 a month for Claude Max or Ultra has bought more usage, not more protection. The agreement that matters is the one in the organisation's settings, and it exists only for Enterprise. Even there, the click that accepts the BAA covers chat, projects and a handful of other features, while skills, connectors, Claude in Chrome and the beta document tools stay outside it. A practice that knows which button is covered is safer than one that only knows its plan name.

Follow me on Instagram@sagnikteaches

Claude plans and BAA routes, one row each

Facts here come from Anthropic's own help articles on HIPAA-ready Enterprise plans and on BAAs for commercial customers, plus its pricing page, checked in November 2026. Anthropic changes these pages often, so confirm each row before you act.

Connect on LinkedInSagnik Bhattacharya

A small clinic will usually fall into one of two positions. Either it is not on Enterprise and therefore has no BAA, or it is weighing the self-serve Enterprise plan against what it would actually use. Anthropic's pricing page lists Enterprise at $20 a seat a month, billed annually, with a 20-seat minimum, plus usage charged at API rates, so the floor is $400 a month ($4,800 a year) before any usage, however few people log in. Either way, nobody should paste patient details into a Free, Pro, Max, Ultra or Team account.

Subscribe on YouTube@sagnikteaches

What the Primary Owner has to do

HIPAA enablement is not a feature you switch on casually. Anthropic describes it as an organisation-level setting that only the Primary Owner of an Enterprise organisation can complete, with no sales or legal cycle needed for eligible organisations.

1. The Primary Owner signs in and opens organisation settings, then Data and privacy, then the HIPAA Compliance section, and clicks Enable.
2. They download and review the BAA, then the Implementation Guide for HIPAA Entities, clicking Next after each. Outside the flow, the guide sits on Anthropic's Trust Center behind an access request.
3. They click "Accept and enable HIPAA". That click is the acceptance of the BAA; there is no separate document to sign and return, and the self-serve BAA is a standard agreement that cannot be modified.
4. A checkmark appears in the HIPAA Compliance section. If it is not there, the organisation is not enabled.

Two points matter for a practice. First, "review before accepting" is not a formality: Anthropic calls the change irreversible for the organisation, and enabling it resets some organisation settings to defaults, so do the reading with your privacy lead, and your legal adviser if you have one. Second, if the person who bought the plan is not the Primary Owner, someone has to be asked. In small organisations the Primary Owner is often the person who set up the account two years ago and may no longer work there, so check who it is before you start.

If you already have an API BAA

If your organisation signed a BAA for Claude API usage before 2 December 2025, Anthropic says that agreement covers the API only. It does not extend to the HIPAA-ready Enterprise plan, so you need a new BAA through your account team. A BAA signed after that date can cover both under a single agreement. Check the date on yours.

What stays outside the agreement after you accept

Enabling HIPAA does not make everything in Claude safe for patient data. Anthropic sorts features into three groups: covered by the BAA, available but not covered, and disabled. Its guidance is to process protected health information only with a feature marked as covered, and to keep features marked otherwise free of it. The authoritative list sits in the Implementation Guide, not in a summary like this one, and it can change.

Anthropic's commercial BAA article gives the feature list in detail. In a HIPAA-enabled Enterprise organisation, chat, projects in chat, artifacts in chat, file creation and code execution (without network access), voice, web search and Research are listed as covered. Skills are off by default and not covered. Connectors, enterprise search and Claude in Chrome are available, but data they send to third parties is not covered, and Claude in Chrome is off by default. Claude Design, Slides and Docs are not available to HIPAA-ready organisations yet, and Cowork in the cloud is not covered.

Further points from Anthropic's pages that a practice should note:

• Claude Code and Cowork. Accepting the BAA does not bring either under it. The Primary Owner must apply the HIPAA configuration to the local mode of each, and Anthropic says that configuration is currently only available to organisations with an Anthropic account team. Anthropic's commercial BAA article says Claude Code via the terminal or Code tab with Enterprise sign-in can be eligible once configured and connected directly to Anthropic, and via a Console API key only with zero data retention. Cloud sessions, the mobile app, VS Code and JetBrains extensions, and gateway or cloud-provider routes are listed as not covered.
• Covered models and retention. Anthropic names certain covered models that need 30-day data retention; in Claude Code they are unavailable while the organisation has zero data retention for Claude Code, unless Anthropic has authorised it.
• Browser agents. Anthropic's safety page for Claude in Chrome says it takes screenshots of the tabs it works in, cannot filter sensitive content out of them, and recommends not using it on sensitive sites. With the BAA not covering its third-party data flows either, a patient portal is no place for it.
• Accounts on other plans. Staff who also use a personal Pro or Max account are outside the agreement however careful they are.

The consequence is plain. The HIPAA checkmark tells you the organisation has a BAA. It does not tell you which button, connector or mode is safe. That is the job of the feature list, and of your own rules for staff.

A six-person dental practice with Max and Team accounts

Say a dental practice with one dentist-owner, one associate, two hygienists, a practice manager and a receptionist wants help with three things: patient reminder wording, treatment-plan letters, and a monthly summary of recall rates. The owner has a personal Claude Max account and the practice manager has Claude Team. What now?

The practice has no BAA today. So letters from notes stay out of Claude until it has one. Self-serve Enterprise would be the route, but the sum is awkward for six people: the 20-seat minimum at $20 a seat means paying for 14 empty seats, $400 a month or $4,800 a year before usage at API rates, to cover one letter-writing task. Compare that with what the practice pays now, roughly $100 a month for the owner's Max plan plus about $40 to $50 for the smallest Team workspace. The honest answer may be to keep Claude for the two non-patient jobs and use a documentation tool that signs a BAA as standard for the letters.

Do-not-paste examples for Claude plans without a BAA

Here are realistic mistakes, each with the safe version. They apply to Free, Pro, Max and Team accounts.

Mistake one: the helpful transcript. A hygienist pastes a recorded consult transcript to tidy it up. It contains the patient's name, a phone number and a diagnosis. Safe version: nothing from the record goes in. Ask instead for a blank structure: "Give me a clean layout for a hygiene visit note with headings for history, findings, advice and follow-up."

Mistake two: the half-anonymised email. "Reply to Mr T about his root canal on Thursday." The initial, the procedure and the day are enough to identify someone in a small practice. Safe version: ask for a reply template with brackets, and fill in the details yourself.

Mistake three: the screenshot. A manager uploads a screenshot of the diary to ask for gaps. The diary shows patient names. Safe version: export only the free slots, or ask for a method for finding gaps and apply it yourself.

Safe prompt (no patient data):
"Write three versions of a text message reminding someone of an appointment tomorrow, under 160 characters, polite, with [time] and [clinic phone] as placeholders. Do not include any health information."

The pattern is the same each time: ask Claude for the shape, then fill the patient detail in by hand inside your own system.

What staff can still do with Claude when there is no BAA

A missing BAA does not make Claude useless to a clinic. It limits the inputs, not the tool. Claude's Projects feature lets you keep a set of instructions and reference files together, and Anthropic's help article says it is available to all users, with Free accounts limited to five projects. A practice can set up a Project holding only non-patient material, such as its own policies, its price list, its opening hours and its tone of voice, and let staff draw on it for everyday writing.

An illustrative set of jobs that need no patient data, with a sample prompt for each:

• Recall and reminder templates. "Write a recall letter template for patients who have not visited in 12 months. Placeholders in square brackets, under 120 words, no clinical claims."
• Job adverts and interview questions. "Draft an advert for a part-time receptionist at a small clinic, with five interview questions about handling anxious callers."
• Policy drafting. "Draft a one-page cancellation policy with a 24-hour notice period and a gentle tone. List three points a lawyer should check."
• Review replies without detail. "Write a calm public reply to a one-star review that says the wait was too long. Do not mention any treatment or confirm the person was a patient."
• Spreadsheet help. "I have columns for month, appointments booked and appointments attended. Write a formula for attendance rate and a short explanation for my practice manager." (Aggregated numbers only.)

The review reply deserves a second look. Even confirming that a reviewer was a patient discloses something, so the safe reply thanks them for feedback and invites them to contact the practice directly, without saying anything about care. Claude will usually follow that instruction if you state it.

A sample output and what you would change

Here is an illustrative reply to the cancellation policy prompt above, trimmed.

Cancellations and missed appointments

We understand that plans change. If you cannot attend, please let us know at least 24 hours before your appointment so we can offer the time to someone else. Missed appointments or cancellations with less than 24 hours' notice may be charged at [fee or percentage].
We will always listen if something unexpected has happened.

What to change: "may be charged" is unclear about who decides, so add "at the practice manager's discretion". The fee placeholder needs your real figure, and the text should say how a patient can pay or dispute it. Finally, ask a lawyer to check that the wording is allowed for your service and your local rules. Claude produces a tidy draft; it does not know what your professional body permits.

A solo therapist, a group practice and a software vendor

A solo therapist. The main value is in session notes and letters, which are patient data. With no route to Enterprise that fits one person, the sensible choice is a documentation tool built for therapists that signs a BAA as standard. Claude stays for the business side: marketing copy, a website FAQ, and invoice templates. The therapist vendor comparison covers that choice.

A medium-sized group practice. With 20 or more staff who would use it, the seat minimum stops being a penalty, and self-serve Enterprise is worth pricing, especially if the group wants a shared assistant that can draft letters from notes. The decision hinges on the Implementation Guide. Ask which features the clinicians would use daily, and check each against the covered list before you sign.

A software vendor building a patient-facing tool. The relevant route is the API BAA, plus your own BAAs with your customers. Anthropic's API documentation says HIPAA readiness does not cover the Console, Claude Code, Amazon Bedrock or beta features, and that the API returns an error if a HIPAA-enabled organisation calls a feature outside the eligible list. This is a different project, and the contract needs a lawyer.

Consumer retention and why it matters here

For individual Claude accounts, Anthropic's consumer terms let users choose whether chats are used to improve its models, and the retention period differs with that choice: Anthropic says chats are deleted from its back-end systems within 30 days, while chats you allow it to use for training may be kept in de-identified form for up to five years. That is a privacy trade-off for a consumer, and it has no bearing on HIPAA. A practice might reasonably think "training is off, so it must be fine". It is not: the question is whether a BAA exists and covers the feature, and on an individual plan it does not. Turning off training protects your content from model improvement; it does not make a vendor a business associate.

A simple decision path

1. Will the task involve information that identifies a patient or could be linked to one? If no, any plan is fine, and the usual business-privacy settings apply.
2. If yes, do you have a signed BAA for this exact product and plan? If no, stop and use a tool that has one, or remove the patient data.
3. If yes, is the feature you plan to use marked as covered in the Implementation Guide? If no, stop.
4. Is the person using it logged in to the covered organisation account, not a personal one? If no, stop.
5. Does your policy say what the output is used for and who reviews it? If no, write that down before you start.

Printed on one page, those five questions replace most of the training a small practice needs. Put them next to the screen where staff work, and revisit them each time Anthropic or any other vendor changes the product list.

Questions to ask Anthropic before you commit

1. Is our organisation eligible for the HIPAA-ready plan on self-serve Enterprise, or do we need the sales-assisted version for what we want (for example Claude Code under the BAA)?
2. Can we see the Implementation Guide before we buy? (Anthropic's Trust Center grants access on request.)
3. Which features that our staff would actually use are covered, available but not covered, and disabled?
4. How do we handle staff who also hold personal Claude accounts?
5. What is the process if we need to leave: data export, deletion and what happens to the BAA?
6. If a team member leaves, what happens to their conversations and projects?

The public list price for self-serve Enterprise ($20 a seat a month, billed annually, 20-seat minimum, plus usage at API rates) is the starting point; usage costs vary with the model and the task, so ask for an estimate based on your expected volume. Keep the written answers with the BAA in your compliance folder.

Policy wording to give staff

Whatever you decide, staff need a rule they can follow on a busy Monday. This wording works for a practice that has no BAA with Anthropic.

Claude and patient information

- Claude accounts at [practice name] are not covered by a business associate agreement.
- Never enter anything that identifies a patient: name, date of birth, address, phone, email, record number, photo, scan, diary screenshot or a combination of details that could identify them.
- You may use Claude for: our own policies, newsletters, job adverts, staff rotas using initials, and general questions with no patient detail.
- If you are unsure, do not paste. Ask [privacy lead].
- If you paste something by mistake, tell [privacy lead] the same day.

If the practice does later enable HIPAA on an Enterprise plan, change the first line, list the covered features by name, and keep the personal-account rule. The AI usage policy for a medical practice has a longer template.

How Claude compares with the other assistants

Seen next to the other assistants a practice might consider, Claude's position is not unusual: consumer and small-team plans have no BAA, and a covered route exists for larger organisations. The details differ, though. Compare what ChatGPT offers by plan, how Microsoft 365 Copilot's agreement treats web search, and what a Google Workspace BAA covers. For a broader view across all of them, the medical office BAA check lays the options side by side.

My practical advice for a small practice is to decide what you want AI to do before you decide which assistant to use. If most of the value is in letters, notes and inbox replies, you need a covered tool; if most of it is in marketing, policy drafting and analysis of aggregated numbers, you may not need a BAA at all. If you would like help sorting that, my AI implementation consultation starts with exactly that list.

Check it worked: a short audit after you decide

Whichever route you pick, run this check a month later. Ask each person which Claude account they use and for what. Confirm that nobody is using a personal account for work. Look at the HIPAA Compliance section in organisation settings, if you have one, and check for the checkmark. Read the feature list again to see whether anything changed. Write the date in your risk assessment. The audit takes twenty minutes and catches most drift.

One final reminder: vendor pages change, and so do eligibility rules. The statements above reflect Anthropic's published guidance at the time of writing. If a vendor salesperson tells you something that differs, ask for it in writing and read the BAA itself before you rely on it. Your legal adviser, not this tutorial, decides what your obligations are.

What practices ask next about Claude and patient data

Can I put patient notes into Claude Pro or Max?

No. Anthropic says individual plans (Free, Pro, Max and Ultra) cannot enable HIPAA, and there is no BAA for them. Paying for a higher tier raises usage limits only. Keep anything that identifies a patient out of those accounts.

Is there a BAA for Claude Team?

Anthropic's HIPAA-ready article says Team plans cannot enable HIPAA, and its commercial BAA article does not mention Team as covered. Treat Team as having no BAA unless Anthropic tells you otherwise in writing, and keep patient data out of it.

Who at my organisation can accept the BAA?

Only the Primary Owner of the Enterprise organisation. Other owners and admins cannot complete the flow. If you are an admin but not the Primary Owner, ask them to sign in and accept. Acceptance is a click-to-accept step, the self-serve BAA cannot be modified, and Anthropic describes the change as irreversible.

Does accepting the BAA cover Claude Code?

Not by itself. Anthropic says enabling HIPAA does not bring Claude Code or Cowork under the BAA. The Primary Owner must also apply a separate HIPAA configuration to their local modes, currently only for organisations with an Anthropic account team, and routes such as cloud sessions and IDE extensions stay uncovered.

Does the BAA make my practice compliant?

No. It covers Anthropic's obligations for the covered features. Your policies, staff training, access controls, risk assessment and rule about which features may handle patient data are still your responsibility.

Further reads

• Is an AI Scribe HIPAA Compliant? A BAA and Consent Checklist Before Go-Live — BAA and consent checks before an AI scribe goes live.
• AI Secretary for Doctors: Referral Letters, Inbox, Insurer Forms — Letters and inbox jobs a doctor can hand to AI.
• HIPAA-Compliant AI Receptionist: A BAA and Retention Checklist — BAA and retention checks for an AI front desk.
• ABA Session Notes With AI: From Raw Session Data to a Draft Note — A practice-owner workflow for turning trial data into a draft ABA note: de-identify or use a BAA tool, a prompt, a worked example and sign-off.
• ChatGPT for Clinicians: Does the Individual BAA Cover Your Practice? — What the individual BAA in ChatGPT for Clinicians covers for each person in a practice, who is left out, and what an organisation-wide agreement needs.
• HIPAA Compliance Policy for a Home Care Agency: Build It With ChatGPT — How to draft each HIPAA policy section with AI using no client data, then the exact items a lawyer and your compliance officer must check.
• Medical Billing AR Follow-Up: Prioritise Aged Claims With AI — Turn an aged AR report into a ranked daily work queue, payer call scripts and consistent notes, using AI on de-identified data only.
• Medical Billing Client Onboarding: Contracts, Access, AI Reports — A medical billing company's onboarding sequence for a new practice, from contract and BAA to access, enrolment checks and the first report.
• Is There an AI Tool for Pharmacists? What US and UK Teams Can Use — Yes: ChatGPT for Clinicians for verified US pharmacists, general business-plan tools elsewhere. What to use them for and what to keep out.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Anthropic help articles HIPAA-ready Enterprise plans, Business Associate Agreements (BAA) for Commercial Customers, What are projects, How long do you store my data and Use Claude in Chrome safely; Claude pricing page; Claude API data retention documentation (all read 4 November 2026).