AI and data safety
MedBrains uses AI for clear-cut operational checks and data quality. Diagnosis, treatment and dose always stay with the clinician, a person approves every action, and patient data is protected before AI sees it.
What our AI is for
In the product today
The assistant reads through the same screens and permissions as the person using it. It cannot open a record that person could not.
When the assistant is given a patient's context, it receives clinical facts such as allergies and active medicines, never the name, UHID or other identifiers.
Mental-health entries the user may not see are withheld from the assistant too.
Anything the assistant drafts carries an AI label and reaches the case sheet only when a clinician inserts it. It never creates or signs an order.
Every lookup the assistant makes is recorded on the audit trail, under the user it acted for.
Policy we hold ourselves to
New AI features are limited to operational checks and data quality. They do not suggest diagnoses, treatments or doses.
No more and no less than the task needs. Every fact in an answer traces back to a field in the record.
If the record does not say it, the AI says so instead of filling the gap.
A clinician or staff member approves every action. The AI prepares; people decide.
Everything the AI drafts is marked as AI-drafted wherever it appears.
Every AI action is recorded in the audit trail as the AI acting on behalf of a named user.
The AI can never see or do more than the person signed in.
Before any AI feature handles patient data
How we check. Every AI feature must pass our test suite for accuracy and data leakage before release: answers checked against the source record for anything added or left out, and every output checked for identifiers that should never be there.
These protections are being built into every AI feature now. An AI feature that handles patient data is released only once they are in place.
Across MedBrains
Each hospital's data is isolated by row-level security in the database itself, not only in the app.
The application connects to the database as a restricted role, so even the app itself cannot step around each hospital's row-level rules.
Every request is checked on the server against the user's role, with per-user extra or denied permissions.
Authenticator-app codes with one-time recovery codes, required by role if the hospital chooses, and single sign-on.
Who saw or changed what, and when.
A full database backup every day, and an hourly check that the latest one landed.
Our hosting runs in AWS Mumbai. Or install MedBrains on your own servers and patient data never leaves the building.
Full details in our commercial terms and privacy policy.
No. It does not diagnose, and it never decides, places or signs an order: diagnosis, treatment and dose stay with the clinician. Today the assistant can write out a prescription a clinician asks for, checked against the patient's allergies and drug safety rules, as text the clinician decides whether to use. New AI features are limited to operational checks and data quality, such as a missing pre-auth document or an unacknowledged critical result.
No. The assistant reads through the same screens and permissions as the person using it, and every lookup is on the audit trail.
No. We do not use patient data to train AI models.
For patient data our standard is AI models hosted in India (AWS Mumbai), with no third-party routers in between. Each hospital's AI provider is set in its own settings.
The AI does not run. You see a message that it could not run safely, and the work carries on without it.
Your hospital does. We process patient data only on your instructions, export it whenever you ask, and return and delete it when you leave.
We do not claim certifications we do not hold. MedBrains is built for ABDM, NHCX and UHI but not yet certified by NHA, and we act as a data processor under the DPDP Act, 2023.
Book a demo
Tell us about your hospital. We'll walk through the screens your teams would use — with your workflows, not a slideshow.
Rather talk now? WhatsApp us or write to hello@medbrains.in.
