Is Claude HIPAA Compliant? Plans That Include a BAA
Only in specific configurations. Anthropic signs the agreement HIPAA requires for its API and for Enterprise plans set up as HIPAA-ready. The Free, Pro and Team plans are not covered, so patient information must not go into them. And a signed agreement is necessary but not sufficient: the practice still owns its side of HIPAA, whichever plan it is on.
This is general information to help you ask better questions, not legal advice. Plan terms change, so confirm the current position with Anthropic before relying on it, and have your compliance officer or healthcare counsel sign off on anything that turns on your specific facts.
What does a BAA have to do with Claude?
Everything. Protected health information (PHI) is health information that identifies a patient — a name next to a diagnosis is enough. A business associate agreement (BAA) is the contract HIPAA requires between a healthcare organisation and any vendor that creates, receives, stores or transmits PHI on its behalf. Typing a patient's details into Claude makes Anthropic that vendor, so no BAA means no PHI.
The HHS guidance on business associates (opens in new window) sets out what the agreement has to contain. Encryption, security reports and a vendor's promise not to train on your data are all worth having, but none of them stands in for the contract.
Which Claude plans can be covered by a BAA?
Two of Anthropic's own offerings: the Enterprise plan in its HIPAA-ready configuration, and the API under the configuration Anthropic specifies, which centres on zero data retention. Claude reached through AWS or Google Cloud is a third route, but there the agreement is with the cloud provider, not Anthropic. Every other plan is out.
| Plan or surface | BAA eligible? | Notes |
|---|---|---|
| Free | No | Not covered |
| Pro (individual) | No | Not covered; the same applies to the other individual plans |
| Team | No | Not covered, despite central administration |
| Enterprise (HIPAA-ready) | Yes | The organisation's Primary Owner must enable it; the BAA is signed with Anthropic |
| API (zero data retention) | Yes | Under the BAA; requires the specific configuration Anthropic sets out |
| Claude on AWS Bedrock | Via the AWS BAA | Covered by AWS's agreement, not Anthropic's |
| Claude on Google Vertex AI | Via the Google BAA | Covered by Google's agreement, not Anthropic's |
The Team row is the one that catches practices out. Team gives you shared projects, central billing and an admin console, and Anthropic's commercial terms apply to it, so it looks like the business-grade option. It is not covered by a BAA, and a practice that rolled Claude out on Team needs to keep patient information out of it entirely.
What does the HIPAA-ready Enterprise setup involve?
A deliberate switch, not a default. The organisation's Primary Owner — the top-level administrator on the account — enables the HIPAA-ready configuration, and the BAA is signed with Anthropic. Some features may be limited or unavailable in that configuration, so check which ones before you design a workflow around them.
Before staff are told they may use it with PHI, settle who administers the account, how people sign in, which connectors to other systems are allowed, and what the written policy says about what may be pasted in. Those are your decisions, and they are where most of the risk sits.
How does the API fit, and what is zero data retention?
The API is how Claude is built into your own software or an automated process. Zero data retention is an arrangement in which Anthropic does not keep your prompts and responses after the request is processed, beyond narrow exceptions it sets out. Not every API feature necessarily runs under it, so confirm each one you rely on.
One gap is worth naming. We have not seen first-party confirmation that Claude Code, Anthropic's coding tool, falls under the BAA, so we treat it as not covered. Keep patient data out of any repository, test fixture or log it can read.
Are Claude on AWS Bedrock and Google Vertex AI covered?
They can be, but not by Anthropic. When you reach Claude through Amazon Bedrock or Google Vertex AI, your contract is with the cloud provider, so it is AWS's or Google's BAA that has to cover the service. Check the provider's list of covered services names the one you use, and sign that agreement before any PHI is sent.
This route suits practices already running on one of those clouds, since the agreement, access control and logging sit alongside everything else there. It is still a build rather than a chat window, so it belongs with custom AI agent development rather than a staff rollout.
What can a practice do with Claude without a BAA?
Plenty, as long as no patient is identifiable. Drafting internal SOPs, policy templates, job adverts, staff training material, general patient-education copy and marketing content are all ordinary work for any plan. The test is the data in each conversation, not the job title of the person typing it.
Be careful with “de-identified”. Removing a name is not enough: HHS's de-identification guidance (opens in new window) lists dates, record numbers, contact details and other unique codes as identifiers too. A pasted clinical note with the name blanked out is usually still PHI. If in doubt, it goes to the covered plan or nowhere.
Is a signed BAA enough to make Claude HIPAA compliant?
No. The agreement makes Anthropic accountable for its side; the practice remains responsible for its own. That means a risk analysis, access controls, staff training, a written policy on what may go in, and the minimum-necessary principle — sharing only the patient information a task actually needs. The BAA is a precondition, not the finish line.
Connections widen the boundary. If Claude is connected to a calendar, CRM or document store, every system PHI reaches needs its own coverage, and Anthropic's agreement does not extend to them. The HHS Security Rule guidance (opens in new window) describes the safeguards expected of the practice itself, and our list of security questions to ask any AI vendor is a practical way to check each one.
What we build, and what we do not claim
Lightray AI is independent of Anthropic: we are not a partner, reseller or affiliate, and there is no government HIPAA certification for software or consultants, so we hold none. What we do is map where patient information would flow, put Claude on the plan that fits the work, keep PHI out of the plans that cannot carry it, and document the setup for your compliance officer to review. They decide what is acceptable; we build to that decision.
Where to go next
If you are setting Claude up for a team, the Claude implementation services page covers plan choice, administration and staff training. For the wider picture of what a clinic can automate around HIPAA, see AI automation for healthcare practices. And if Claude is going to sit alongside an automation platform, the guide to Zapier and HIPAA explains why each tool in the chain needs its own agreement.
Common questions about Claude and patient data
Is Claude HIPAA compliant out of the box?
No. No software is HIPAA compliant on its own, and Claude is no exception. Patient information can only go into Claude under a signed business associate agreement with Anthropic, on a HIPAA-ready Enterprise plan or the API configured for zero data retention. Even then, the practice still carries its own HIPAA obligations.
Does Anthropic sign a BAA?
Yes, for HIPAA-ready Enterprise plans and for the API under the configuration Anthropic specifies. The Free, Pro and Team plans are not covered, whatever they are used for. Ask Anthropic for the agreement itself, and check that it names the product and configuration you will actually use before any patient data goes in.
Can a small practice use Claude for patient-adjacent work?
Yes, within limits. De-identified drafting, internal SOPs, policy templates, staff training material and general patient-education copy are fine on any plan, provided no patient details go in. Anything that touches protected health information, including a pasted note or a referral letter, needs the HIPAA-ready Enterprise plan with the BAA or the API.
Does Anthropic's BAA cover Claude on AWS Bedrock or Google Vertex AI?
No. On those platforms your agreement is with the cloud provider, so it is the AWS or Google business associate agreement that has to cover the service, not Anthropic's. Check that the provider's list of covered services names the one you use, and sign its agreement before sending patient data.
Is Claude Code covered by Anthropic's BAA?
We have not seen first-party confirmation that it is, so we treat it as not covered. Keep patient data out of any codebase, test fixture or log that Claude Code can read, and ask Anthropic in writing if your team needs it in scope.
Does Lightray AI hold a HIPAA certification or an Anthropic partnership?
No to both. There is no government HIPAA certification for software or consultants, and Lightray AI is independent of Anthropic. What we offer is a documented data flow, Claude set up on the plan that fits the work, and a build your compliance officer can review before anything goes live.
Free consultation
Tell us what is costing you the most time
Describe the process. You get a straight read on whether it is worth automating, roughly what it would take, and what we would tackle first — on a call that costs nothing.
- Under an hour, free, and no obligation
- Fixed price per stage, agreed before any work starts
- You own everything we build, documented and in your accounts
- If you do not need us, we will say so on that call
The form is the fastest way to reach us — we reply from a real person, usually within one business day.