AI SCRIBE · SECURITY & COMPLIANCE
Security Built for PHI, Not Bolted On After the Fact
Every design decision in AI Scribe — from how audio is captured to how long it's retained — starts from the assumption that it's handling protected health information. Here's exactly how that works.
COMPLIANCE & STANDARDS
What "secure" needs to mean for clinical audio
AI Scribe touches some of the most sensitive data an organization holds — live patient conversations. The security model is built around three commitments that hold regardless of deployment size.
Minimum necessary retention
Raw audio is kept only as long as needed to generate and validate a note, on a configurable retention window, then permanently deleted — not stored indefinitely by default.
No model training on customer data
Patient audio, transcripts, and generated notes are never used to train shared or third-party models. Data is processed solely to deliver the documentation service for that customer.
Full auditability
Every access, generation, and write-back event is logged, timestamped, and attributable to a specific user and encounter — built for HIPAA audit and breach-investigation requirements from day one.
How data is protected at each stage
PHI moves through several stages between capture and chart. Each one has its own protection layer.
Capture
Audio is encrypted immediately at the point of capture (device or browser) before transmission — it is never sent or stored unencrypted, even briefly.
Transit
All data in transit — audio, transcripts, structured notes, API calls — is protected with TLS 1.2 or higher between every service boundary, including EHR write-back.
Processing
Transcription and note generation occur in isolated, access-controlled processing environments. PHI is never logged in plaintext application logs or exposed to engineering tooling used for debugging.
Storage
Any data persisted (transcripts pending review, generated notes) is encrypted at rest with AES-256, with encryption keys managed separately from the data they protect.
Deletion
Raw audio is deleted automatically once it falls outside the configured retention window, following a documented, auditable deletion process rather than manual cleanup.
Access control model
Access to PHI is scoped by role and by need, not granted broadly by default.
Role-based access control
Clinicians, administrators, and support staff each have distinct permission scopes — a support engineer troubleshooting an integration issue does not have the same access as a clinician reviewing a note.
OAuth 2.0 / SMART on FHIR scopes
EHR integration access is limited to the specific FHIR resources required for note write-back — not broad, unscoped access to the patient record.
Multi-factor authentication
Enforced for all administrative and clinical user accounts accessing the platform directly, in addition to any MFA already enforced at the EHR or SSO layer.
Session and device controls
Session timeouts and device-level controls limit how long an authenticated session remains active, reducing exposure from an unattended device.
Deployment models
Different organizations have different constraints around where data can live. AI Scribe supports more than one deployment posture.
| Model | What It Means | Best Fit For |
|---|---|---|
| Managed cloud (multi-tenant) | Hosted by Peerbits with logical tenant isolation, encrypted storage, and standard retention policies | Most clinics, telehealth groups, and mid-size practices |
| Single-tenant cloud | Dedicated infrastructure and isolated data store for one customer, same security controls | Larger health systems with stricter data-isolation requirements |
| Private VPC / customer cloud | Deployed into the customer's own cloud environment (e.g. their AWS or Azure tenant) | Enterprise health systems with existing cloud governance requirements |
BUSINESS ASSOCIATE AGREEMENT
A signed BAA is a standard part of onboarding for every healthcare customer, defining how PHI is handled, stored, and protected in line with HIPAA requirements — not a separate negotiation.
WHAT WE ASK IN RETURN
Security is shared. Customers are responsible for their own user access hygiene (strong passwords, MFA where available, prompt offboarding of former staff) — the platform enforces controls, but organizational practices still matter.
Frequently asked questions
Yes. AI Scribe is built on a HIPAA-compliant architecture with encryption in transit and at rest, role-based access controls, full audit logging, and a signed Business Associate Agreement available to every covered entity or business associate customer.
No. Customer audio, transcripts, and generated notes are not used to train shared or third-party models. Data is processed solely to deliver the documentation service for that customer.
Raw audio is retained only as long as needed to generate and validate the structured note, on a configurable retention window set per customer, after which it is permanently deleted.
Yes. A signed BAA is provided as a standard part of onboarding for every healthcare customer.
The platform is built with SOC 2-aligned controls and undergoes regular third-party security assessments. Detailed security documentation, including architecture diagrams and control descriptions, is available upon request for your compliance and legal review.
Yes. Private VPC and customer-managed cloud deployments are supported for organizations with strict data residency or governance requirements. The same security controls apply regardless of deployment model.
Have more questions?
Ask our expertsNeed this reviewed by your security or compliance team?
We can share detailed architecture documentation, our standard BAA, and answer questionnaire requests directly — happy to get on a call with your security team.
Request security documentationKnowledge hub
Expert insights on healthcare security, HIPAA compliance, and clinical data protection.
