GOVERNED AGENT RUNTIME FOR HEALTHCARE

Deploy AI agents in workflows where PHI cannot leak.

For US payers and providers under HIPAA + HITECH + state insurance regulation, and EU life-sciences institutions under EU AI Act Annex III (medical-device high-risk). PHI stays inside your VPC or air-gap deployment; every agent action authorized against policy before execution; audit trail signed and offline-verifiable.

Request a demo → Read the technical brief →

Why this matters now

Healthcare AI deployment is at an inflection point that regulation is not making easier. HIPAA administrative safeguards (§164.308) predate AI agents by two decades; state insurance regulators are catching up in real-time; and for European life-sciences institutions, the EU AI Act's medical-device high-risk provisions apply from December 2027.

The specific defect that stalls healthcare AI pilots is not model quality — it is the inability to prove, to a HIPAA compliance officer or a state DOI examiner, that Protected Health Information stayed inside the boundary you promised the patient it would, at every moment of the agent's execution. A chat transcript does not prove that. A downstream audit log written after the fact does not prove it either.

Orkena is designed for exactly that proof: PHI-labeled data never crosses a boundary the policy does not permit, and every decision to permit or deny is recorded in the same transaction that made it.

How Orkena maps to the healthcare regime

Regulation	What it requires	What Orkena provides
HIPAA §164.308 (administrative safeguards)	Access authorization, workforce clearance, security awareness	Scoped agent credentials via KMS; per-user role-based access with RLS; SSO integration; audit trail of every access decision
HIPAA §164.312 (technical safeguards)	Access controls, audit controls, integrity controls, transmission security	Hash-chained ledger of every PHI-adjacent action; Ed25519 signed anchors detecting any post-facto modification; TLS 1.3 in transit; envelope encryption under your KMS at rest
HITECH breach notification	Ability to determine what PHI was involved in a breach	Deterministic replay + evidence bundle export scoped to any incident window; auditor runs `verify.py` on their own machine
State insurance model laws (e.g. NY DFS 500, CA DOI)	Cybersecurity program, incident response, third-party risk	Kill switch multi-level; DR posture; vendor DPA; sub-processor list on the trust centre
EU AI Act Art. 12 (high-risk logging)	Comprehensive record-keeping for medical-device AI	Per-tenant ledger; anchors; bundle export
EU AI Act Art. 14 (human oversight)	Human oversight controls for high-risk AI	Four-eyes approval gates; SLA + escalation; conflict-of-interest blocking

Honesty note: Orkena is not HIPAA certified on your behalf. It provides runtime controls designed to support HIPAA administrative and technical safeguards. Your HIPAA compliance officer, Privacy Officer, and Security Officer interpret them for your specific covered-entity or business-associate posture.

Four healthcare use cases Orkena runs today

1. Prior-authorization assistance (payer).

Agent evaluates member requests against policy, proposes decision, escalates to human medical director. Policy denies external tool calls when PHI is present in payload; four-eyes required for any denial that could affect member care.

2. Claims triage (payer).

Agent classifies incoming claims, proposes routing (auto-adjudicate / human review / SIU). Every routing decision recorded with the input, the policy that fired, and the classification confidence. Deterministic replay for any audit request.

3. Care-plan drafting assistance (provider).

Agent proposes care plan based on chart data (retrieved from EHR via MCP). Draft never reaches a system of record without clinician approval (HITL). Policy denies model calls with certain data-label combinations (e.g. genomic data flagged for research-only use).

4. Regulatory documentation assistance (life sciences).

Agent drafts regulatory submissions from source documents; every draft is a versioned artifact; changes require human approval; complete audit trail for FDA / EMA scrutiny.

For each: authorize before execution, evidence in the same transaction, deterministic replay, evaluation-gated updates.

The demo you will see

PHI-labeled data denial. An agent attempts to call an external tool with content flagged as PHI. The policy engine denies the tool call before it fires. The denial is recorded with the specific data-label match that triggered it.
Four-eyes for a member-affecting decision. A prior-authorization denial requires two clinicians to approve. The initiator cannot approve their own. The Face-ID mobile approval closes the loop in seconds.
Air-gap parity demo. Orkena runs the same agent workflow in an air-gapped deployment (no outbound connectivity) — local embedder, local classifiers. PHI never leaves the network you control.
Evidence bundle verified offline. An auditor at a state DOI examines a case-triage decision. The evidence bundle is verified on their machine with verify.py. The bundle is tamper-evident; any modification returns FAILED.

What Orkena is not for you (in healthcare)

Not a HIPAA covered entity. Orkena provides the runtime; your compliance posture is your own. Where Orkena processes PHI on your behalf (the managed, region-pinned tier, on the roadmap), Orkena acts as a business associate and signs a BAA. In customer VPC and air-gapped deployments Orkena operates nothing and has no access to your data, so a BAA does not apply.
Not an EHR replacement. Orkena orchestrates agents that interact with your EHR (via MCP or existing FHIR APIs); it does not replace your EHR.
Not a clinical decision support system for FDA-cleared use. Orkena is the substrate; clinical judgment stays with clinicians, and any clinical-decision use case requires your own FDA clearance / CE mark process.
Not shipping SOC 2 Type II today. Readiness in progress; audit planned. Design partners in healthcare receive control mappings ahead of the audit report.