Originally published on Nirmitee.io: HIPAA Compliant AI Medical Scribe: How to Check One, and How to Build One
"Is it HIPAA compliant?" is the first question most buyers ask about an AI medical scribe. It is a real requirement. It is not the one that sinks these projects.
There is no official HIPAA certification for software, and HHS does not certify vendors. So "we are HIPAA compliant" is not an answer. A HIPAA compliant AI scribe is one whose vendor signs a business associate agreement (BAA) covering the exact plan you use, and whose product enforces consent, access limits, audit logs and deletion of recordings in code.
Ten questions to ask every vendor before a pilot
A good answer is specific and backed by a document or a test.
- Will you sign a BAA for the plan we are buying? Many self-serve plans do not include one.
- Which subcontractors receive audio, transcripts or prompts? Look for a named list, each covered by a BAA.
- Is audio or text used to train your models? A good answer is no by default, or opt-in in writing.
- How long do you keep audio, transcripts and drafts? A retention period you set, with deletion you can verify.
- How do you record patient consent? Per participant, before recording starts.
- Who inside your company can access our data? Role-based, logged and time-limited.
- Can we see the audit log? You need to answer "who opened this session".
- Where is data stored and processed? Named regions, in the contract.
- Is every note a draft until a clinician signs? No auto-signing, with edits recorded.
- What happens to our data when we leave? Deletion within a stated period, confirmed in writing.
Check the plan tier every time: a free or individual plan is where the BAA is most often missing.
Why a BAA is not enough
An AI scribe can satisfy every HIPAA rule and still write a progress note that invents a symptom, suggest a diagnosis code that does not exist, record a session in a state that requires everyone's consent, or say something to a distressed client that only a clinician should say. None of those is a HIPAA violation. All of them end a pilot.
The cost is rarely a fine. It is a clinician signing a note with a symptom the client never described, a supervisor who stops trusting every draft after one bad one, and a sales cycle that stalls on a compliance officer's questions. The original guide walks through these risks in the order they occur, from booking to deletion.
The code around the model is the product
Every AI scribe has a model, a prompt and a screen. The ones that survive review share one design decision: anything that must always be true is enforced outside the model.
- Consent gates capture. If consent is not on file for every participant, recording does not start. The source lists ten all-party consent states, including California, Florida and Washington. Illinois requires written consent before AI records or transcribes a therapy session.
- Everything is a draft. Only a clinician promotes it, edits are recorded, and signed notes refuse the AI.
- Codes are checked, not requested. Each suggested diagnosis code is checked against the ICD-10 catalog, and unknown codes are dropped.
- Crisis language gets a fixed, approved response that the model never generates.
- Same clinic is not same access. Only the treating team, the assigned supervisor and the clinic admin can open a session.
- Purge the payload, keep the evidence. Recordings are deleted on schedule; usage and audit records without patient information stay.
Is the business case real?
Yes, and more modest than the marketing. A JAMA study across five academic health systems and 8,581 clinicians found about 16 fewer minutes of documentation per eight hours of scheduled patient time, and only about a third of adopters used the tool in most visits. A separate 30-day study of 263 clinicians saw burnout fall from 51.9% to 38.8%.
Time savings arrive only when clinicians trust the drafts enough to use them.
Build, integrate or let customers bring their own
For a software vendor, there are three routes:
- Build it into your product when documentation is central and you want the margin and the data model.
- Integrate a scribe vendor's API when you need it this year. You still own consent, access, the note lifecycle, retention of what you store, and the vendor's BAA.
- Let customers bring their own when documentation is peripheral. You own very little, and get very little of the value.
The vendor cannot enforce your consent rules or access model, because those live in your product. The full guide covers transcription APIs offered under a BAA and a twelve-question evaluation checklist.
Nirmitee.io builds healthcare AI into EHR and practice software, with consent, access and audit enforced in code. For the full buyer's check and build guide, read the original: HIPAA Compliant AI Medical Scribe on Nirmitee.io.
Dinesh Thorat is Healthtech Growth Partner at Nirmitee.io.
Originally published at nirmitee.io.
Top comments (0)