Data Handling
Where your data lives, how it's isolated, and what we don't do with it.
Tenant isolation, PHI handling, model-training boundaries, and data residency — the topics every healthcare security review asks about.
Last reviewed:
Tenant isolation
Every Synaptis customer operates inside a logical tenant boundary. PHI, customer-uploaded documents, model context, and audit logs are partitioned per-tenant at the data layer. Encryption keys are per-tenant and managed via cloud KMS. There is no shared mutable store crossing tenant boundaries.
- Per-tenant encryption keys (cloud KMS)
- Per-tenant audit log streams
- Per-tenant rate limiting and quota
- Per-tenant model context — no shared cache
Model training
Customer data is never used to train, fine-tune, or otherwise improve any model that serves other customers. This is contractually committed in every BAA and MSA, and architecturally enforced at the data plane.
- No cross-tenant training, ever
- No model improvements from customer prompts
- Anonymized aggregate metrics (e.g., scribe accuracy benchmarks) require explicit opt-in via an addendum
PHI handling
All PHI is processed inside the Synaptis HIPAA data plane. PHI in transit uses TLS 1.2+; PHI at rest is encrypted AES-256 with per-tenant keys. PHI is not written to debug logs; redaction is enforced at the logger boundary.
- PHI processed only inside HIPAA-eligible cloud regions
- PHI redaction enforced at the logging boundary
- PHI retention configurable per tenant; default 7 years to match common HIPAA expectations
- PHI export available to tenant administrators on request
Data residency
Default deployment is US-region. EU-region hosting is available on request and is contractually committed where activated. Cross-region replication is opt-in only.
Data subject requests
For GDPR/CCPA data subject requests, customers route requests through their tenant administrator. Synaptis fulfills the underlying technical operations on the tenant administrator's instruction. See /privacy for the customer-facing process.