Why Standard Claude Is Not HIPAA Compliant
HIPAA compliance comes down to a legal relationship, not a feature. A vendor is only HIPAA compliant for your practice's use once they've signed a Business Associate Agreement (BAA) with you, accepting legal responsibility for how they handle protected health information (PHI) on your behalf. The free version of Claude, and the standard Claude Pro subscription, do not include a BAA. That means entering a patient's name, diagnosis, or any identifying health detail into a regular Claude conversation is a HIPAA violation, even if the intent was simply to save time drafting a note.
This is easy to miss because Claude, like most AI chat tools, feels like a private conversation. It isn't, legally speaking. Every message sent is processed on Anthropic's servers, and without a BAA, there's no contractual protection covering how that information is stored, used, or secured.
When Claude Can Be Used Under a BAA
Anthropic does offer BAAs, but through specific enterprise and API-level agreements, not the consumer product most individuals use. This generally means working directly with Anthropic's business offerings, or using a platform that has its own BAA with Anthropic and extends that coverage to your practice.
Even once a BAA is signed, a few things still need to hold up for genuine compliance. Data needs to be encrypted in transit and at rest. Access should be limited to staff who need it, with logs showing who accessed what information and when. There needs to be a clear, documented policy on how long data is retained and how it's deleted. A BAA handles the legal side, it doesn't automatically mean the technical setup is secure, that depends on how the tool is actually configured.
What a Compliant Setup Actually Requires
Before using Claude, or any general AI tool, for anything touching patient information, a practice should be able to answer these clearly: Is there a signed BAA with the vendor covering this specific use? Is data encrypted in transit and at rest? Is there an audit trail of who accessed what patient data and when? Is there a written data retention policy, and does the vendor follow it? Are staff trained on what can and can't be entered into the tool, since a compliant backend doesn't prevent misuse by someone who doesn't know the rules?
If any of these answers is unclear, the safer assumption is that the tool isn't ready for patient data, regardless of how useful it is elsewhere.
A Safer Path: Purpose-Built Healthcare AI
General AI assistants like Claude are designed to be broadly useful, which is exactly why making them compliant for a specific, high-stakes use case like healthcare takes extra setup. Tools built specifically for healthcare, the kind we implement through healthcare AI consulting, come with the BAA, encryption, and access controls already built in, rather than needing to be configured around a general-purpose product after the fact.
This matters most for anything clinically documented or patient-facing, like AI scribe tools transcribing exam room visits, or AI patient scheduling systems handling appointment history. These are exactly the cases where a general consumer tool without a BAA creates real legal exposure.
What to Do If Your Team Is Already Using Claude
If staff have already been using standard Claude with any patient information, the first step is to stop and review what was entered, since that may need to be addressed as part of a compliance review. From there, a clear written AI use policy, covering which tools are approved, what information can and can't be entered, and who to ask when unsure, closes the gap going forward. Our free AI readiness assessment is a good place to start if you're unsure where your practice currently stands.

