MedicDraft Security
Active Tenant
PHI processing is blocked. This demo remains blocked until BAAs, tenant-isolated storage, KMS, audit logging, and required AWS HealthScribe settings are complete.
AWS Readiness Checklist
These controls must be complete before real audio, transcripts, reports, QA findings, screenshots, or patient identifiers are allowed into MedicDraft.
BAA gate is active. Real PHI remains blocked.
Execute and document BAAs with AWS and every ePHI-touching vendor before enabling PHI workflows.
Audio, HealthScribe output, transcripts, reports, screenshots, and exports need tenant-isolated encrypted storage.
Create private S3 storage with block public access, SSE-KMS, tenant-scoped prefixes or buckets, access logs, and lifecycle retention.
Customer-managed KMS encryption is required for PHI-touching storage and HealthScribe workflows where supported.
Create a customer-managed KMS key with least-privilege key policy, rotation, CloudTrail monitoring, and tenant encryption context.
Reports, transcripts, QA findings, feedback, users, audit references, and tenant settings need a BAA-covered database with tenantId on every PHI record.
Choose and configure a BAA-covered database, add tenant-scoped schema, and block PHI writes without tenantId and audit events.
AWS IAM and app RBAC both need least-privilege boundaries before PHI routes or HealthScribe jobs are enabled.
Connect production auth with MFA, server-side session checks, tenant-aware roles, and AWS roles scoped to tenant storage paths.
AWS API activity and app-level PHI actions need audit trails before protected workflows can run.
Enable CloudTrail/CloudWatch destinations and implement immutable app audit writes for create, view, edit, generate, export, admin, and delete events.
HealthScribe remains blocked until environment, BAA, tenant, KMS, IAM, storage, and audit gates are complete.
Keep the server-side HealthScribe route fail-closed until every PHI readiness item is complete and reviewed.
PHI mode is disabled. Sandbox must remain synthetic-only.
Do not enable PHI mode until legal, auth, storage, KMS, IAM, audit, database, HealthScribe, retention, and incident-response controls are complete.
AWS region selected for HealthScribe and tenant storage.
Tenant-isolated encrypted S3 bucket for source audio.
Tenant-isolated encrypted S3 bucket for HealthScribe output.
Customer-managed KMS key for an added HealthScribe encryption layer.
Least-privilege role allowing HealthScribe access to tenant audio/output paths.
Central audit/log destination for PHI-touching workflow events.
Every PHI-touching table, object path, audit event, model artifact, and API request must include tenantId.
Enforcement: Reject any PHI workflow request that does not carry a tenantId matching the authenticated user's tenant.
Written agreement must state that department recordings, reports, protocols, and model artifacts remain private to that department.
Enforcement: Keep training/export workflows disabled until the tenant data-use terms are complete.
Provider, QA, admin, and owner roles need separate permissions for recording, review, export, rules, and reference management.
Enforcement: Server-side permission checks before every PHI read, write, export, deletion, and admin mutation.
PHI-touching actions must write an audit event before or during the protected action.
Enforcement: Fail closed when audit logging is unavailable.
Each department must define retention windows for audio, transcripts, generated narratives, protocol sets, and model artifacts.
Enforcement: Retention policy is required before storage activation.
One department's data must not train, enrich, or evaluate another department's model or retrieval index.
Enforcement: Tenant model profile and retrieval index IDs are stored on the tenant config and must match every AI request.
Creates dictations, reviews own generated narrative, resolves missing information, and copies ESO narrative.
Future role for reviewing submitted charts, unresolved issues, alert calls, and trends.
Manages department templates, protocols, rules, users, and security readiness visibility.
Controls tenant-level settings, PHI gates, vendor configuration, and production security posture.
No real PHI may be processed until BAAs are signed with AWS, hosting, database/storage, AI, and any support vendors.
Tenant configuration keeps PHI storage disabled until legal/vendor readiness is explicitly changed.
Every recording, transcript, report, reference file, audit event, and model artifact must be scoped to a tenant.
Tenant config and future data schemas require tenantId as the first access-control boundary.
Users must authenticate into one organization and receive role-based permissions for provider, QA, admin, and owner workflows.
Architecture reserved for tenant auth provider, MFA, session controls, and role checks before PHI routes are enabled.
Audio, transcript, report, protocol, and HealthScribe outputs must use encrypted tenant storage and customer-managed keys where supported.
HealthScribe configuration requires a KMS key ARN before PHI processing can be enabled.
All access, creation, transcription, generation, copy, export, admin, and deletion actions must create audit events.
Audit event contract is defined before real storage is added.
Each tenant must have documented retention windows and deletion procedures for recordings, transcripts, reports, and model artifacts.
Retention controls are listed as required architecture before database/object storage activation.
Department data cannot train another department's model or vendor foundation models without written approval.
SMFR data may only be used for the SMFR tenant model or retrieval index unless a future written agreement explicitly allows another use.
All HealthScribe jobs must run through a server-side adapter that enforces tenant, BAA, KMS, bucket, role, and audit checks.
Adapter scaffold blocks real PHI while demo mode and missing production environment requirements remain active.
Legal permission to use HIPAA-eligible AWS services with ePHI.
Create transcripts and clinical documentation artifacts from dictation audio.
Store encrypted audio submitted for HealthScribe processing.
Store encrypted transcripts, generated note output, and service artifacts.
Encrypt audio, transcript, output, and future report/model storage.
Allow HealthScribe to read input audio and write output for a tenant workflow.
Record access, generation, export, admin, and deletion events.
AWS HealthScribe will be a server-side integration only. The browser should never receive AWS credentials, KMS keys, bucket names for PHI objects, or direct cross-tenant storage paths.