Building the Healthcare Brain Built for the highest-trust healthcare environments.
From sovereign-deployable architecture to SOC 2 Type II, ISO 27001, and HIPAA-by-design — security is foundational, not bolted on.
The compliance posture.
HIPAA by design
Privacy and Security Rule compliance is architectural. PHI partitioning by tenant. Audit logs on every action. BAA standard with every customer.
SOC 2 Type II
Type II report covering security, availability, processing integrity, confidentiality, and privacy. Available under NDA.
ISO 27001
Information security management system certified. Annual surveillance audits. Statement of Applicability available under NDA.
AES-256 · TLS 1.3
At rest and in transit, with customer-managed keys available for Sovereign deployments.
Sovereign & ATO-ready
Aether One™ Sovereign aligns with FedRAMP requirements and is deployed today inside a CMS Medicare environment.
Continuous testing
Static analysis (SonarQube), dependency scanning, third-party penetration tests, bug bounty (planned).
Security as architecture, not policy.
Some controls are policies a vendor asks you to trust. The ones below are guarantees enforced by architecture — auditable in code.
| Guarantee | How it's enforced |
|---|---|
| No PHI in cross-tenant model training | Data partitioning at the storage layer. Models retrain only on curated, de-identified datasets with explicit, written customer authorization. |
| Sovereign means sovereign | Aether One™ Sovereign runs the full agent stack inside customer perimeter, anchored on Genzeon Platforms' patent-protected Knowledge Containment Architecture. No model weights phone home. No telemetry leaks reasoning. |
| Audit log integrity | Every action logged, timestamped, signed. Reconstructable from on-premise logs alone — no cloud roundtrip required to answer a CMS audit. |
| Mandatory human review on adverse decisions | Architectural, not policy. Agent 871 (Non-Affirm Research) routing is hard-coded into every clinical agent's graph. |
| Threshold cryptography for hybrid deployments | For controlled-egress Sovereign, no single domain holds keys to compromise either model integrity or PHI confidentiality (PA-SEC). |
Open-weight models, published.
Healthcare AI components — especially the ones handling PHI — should be inspectable rather than trusted. The CliniGuard family is our open-weight contribution to that posture: clinical NER models published on Hugging Face under Apache-2.0, with named base models and reproducible architectures. Privacy officers, security teams, and external auditors can review the actual primitive, not a marketing claim about one.
De-identification, open-weight.
20 PHI/PII categories aligned to HIPAA Safe Harbor. Bio_ClinicalBERT base, Apache-2.0, headline F1 0.97 (micro). Recommended for human-in-the-loop pairing in high-stakes workflows.
Clinical extraction, open-weight.
15 vital-sign and measurement categories across nursing notes, ED triage, and progress notes. Bio_ClinicalBERT base, Apache-2.0. The kind of primitive structured clinical pipelines are built on.
Need our SOC 2 report or security questionnaire?
Available under NDA. We'll typically turn around standard questionnaires (HECVAT, CAIQ, SIG) in 5 business days.