Healthcare · HIPAA · AI Governance
The gate every AI query passes through before it touches patient data.
Self-hosted on your hardware. Every AI / agent / automated read of an EHR record routed through the gateway produces a cryptographically signed Decision Receipt. Mass-read patterns trigger human approval. Tampered vendor code is designed to flag within hours of compromise, not the 90 days that NYC Health + Hospitals lost.
The four threats SynOI is designed against
The failure modes turning healthcare breaches into million-person disclosures.
Third-party vendor compromise
NYC H+H (1.8M people), SolarWinds, Kaseya
SynOI Supply Chain Guard: state-divergence detection on each package update (Supply Chain Guard is in active development; public beta Q1 2027). Flags new postinstall scripts, fresh egress domains, cross-package hash collisions.
90-day undetected dwell time
Average dwell time for healthcare breaches: 280 days (Ponemon 2024)
Continuous integrity heartbeat (active): modified vendor code flags degraded:true on every receipt within hours of compromise. Detection collapsing from months to single-digit hours is a modeled improvement, not yet benchmarked against a live incident.
Mass exfiltration via AI / agent paths
Compromised credentials used to mass-query EHR through the legitimate API
Risk Policy + human approval (HITL, active) + mass-read rate-limit (in-flight, tuned per customer during pilot). 'Read 100 patient records in 5 minutes' triggers human approval via SMS / Slack / system tray. Attacker has to compromise the human approver too.
Forensic vacuum during disclosure
NYC H+H disclosed 90 days late with partial forensic claims; class-action exposure inflated by uncertainty
Cryptographically signed receipts (active). Bitcoin and sigstore Rekor anchoring: planned, Q4 2026. After breach: precise, tamper-evident answer to 'which records did the attacker access'.
HIPAA Security Rule mapping
SynOI as auditable evidence for HIPAA technical safeguards.
§ 164.312(a)(1)
Access Control
Risk Policy gates high-risk / high-blast-radius reads through human-in-the-loop (HITL) approval; per-action role + tenant enforcement; cryptographic non-repudiation of each access decision it governs
§ 164.312(b)
Audit Controls
Signed Decision Receipts on every AI / agent action routed through the gateway (active); hybrid Ed25519 + ML-DSA-65 cryptographic signature; counter-signed by SynOI; Bitcoin + Rekor anchoring planned Q4 2026; exportable as auditor evidence pack
§ 164.312(c)(1)
Integrity
Manifest-bound code attestation; receipts signed under the manifest hash so an attacker who modifies the gateway cannot produce a non-degraded receipt
§ 164.312(d)
Person or Entity Authentication
Human-in-the-loop (HITL) workflow records cryptographic identity of the human approver on every high-risk decision it gates; SMS / Slack / mobile push / system tray channels supported
§ 164.308(a)(1)(ii)(D)
Information System Activity Review
Auto-tagged compliance receipts feed directly into review pipelines; per-framework export (NIST AI RMF, SOC 2, ISO 42001) without manual mapping
This mapping is offered for orientation, not legal opinion. Your compliance team should review SynOI's evidence properties against your specific OCR risk posture.
60-day pilot scope
Pilot details. Zero cost. No commitment.
Phase 1 · Week 1-2
Setup
- ▸Pick one AI tool: Epic Cosmos, internal LLM assistant, agent
- ▸SynOI Gateway installed on your hardware
- ▸Routes through your existing AI provider keys
- ▸Observe mode: no actions blocked yet
Phase 2 · Week 3-6
Observation
- ▸Receipt corpus accumulates
- ▸Risk policy baseline learned from real traffic
- ▸Tune mass-read thresholds against your actual volumes
- ▸Weekly 30-minute sync with our team
Phase 3 · Week 7-9
Evaluation
- ▸Receive HIPAA + NIST AI RMF + SOC 2 evidence pack
- ▸Cryptographically verifiable; auditor-ready
- ▸Decision: continue, expand to additional AI tools, or step back
- ▸Pricing conversation only if you continue
Predict before you deploy
Test every policy change against your last 90 days of receipts.
A CISO should never flip a new HITL rule into production without knowing what it breaks. The portal's Policy Preview takes a candidate risk.policy, replays it against your historical receipts, and reports exact numbers: how many actions would now be denied, how many would newly require HITL, which workflows are affected, who the top callers are.
Direct analog to Forward Networks' Forward Predict (network change verification, launched May 2026; IDC reports $14.2M average annual customer benefit). Same predict-before-deploy property, applied to AI governance.
Policy change preview (last 30 days, 142,318 receipts):
Existing policy result:
deny: 12
require_approval: 47
allow: 142,259
Proposed policy result (mass-read HITL added):
deny: 12
require_approval: 189 (+142) ← new HITL fires
allow: 142,117 (-142)
Workflows affected:
Mass-read patient records:
action_type: patient.read
receipt_count: 142
top_caller: epic-cosmos-llm (138)
change: now_requires_hitl
Recommend: notify epic-cosmos-llm team before activating.Cryptographic posture
Verifiable today. Quantum-resistant today.
Receipts are signed with hybrid Ed25519 ∥ ML-DSA-65. Every new receipt carries BOTH a classical Ed25519 signature AND a post-quantum signature under NIST FIPS 204 (the standard finalized August 2024). Both must verify. Auditors verify Ed25519 with any standard library and ML-DSA-65 with any FIPS 204 library, no SynOI runtime required for either path. Source code is public on GitHub; the verification mechanism is auditable.
Healthcare retention windows reach 7-25+ years. Receipts signed today are quantum-resistant, not on a roadmap. The hybrid model also gives cryptographic agility: a future bug in either library cannot compromise an authentic receipt because the OTHER signature is still required. Pre-Sprint-11 Ed25519-only receipts remain verifiable under the classical path alone.
We follow Kerckhoffs's principle: a cryptographic system should be secure even when everything except the key is public. Hiding our integrity-check code wouldn't raise determined-attacker cost meaningfully and would make us look untrustworthy to your security team. The security is in the keys, the hybrid Ed25519 + ML-DSA-65 signatures, and the multi-party signing flow.
Deployment
Self-hosted is the recommendation for HIPAA.
Self-hosted
RecommendedGateway runs on your hardware. Uses your existing Anthropic / OpenAI / Azure keys via per-request headers; SynOI never stores them. PHI never traverses SynOI infrastructure. The default for HIPAA-bound customers.
SynOI-hosted cloud
Not live yet · not the rec for HIPAANot live in production; on the roadmap as a paid tier for non-regulated workloads. Listed for completeness; for healthcare we strongly recommend self-hosted, which is what production deployments run on today. OAuth subscription mode will be refused in hosted (Anthropic ToS) once it ships; API-key mode is fine either way.
For Utah-based healthcare systems
Local. Same time zone. Founder-direct.
SynOI is headquartered in Utah. Founded by Joshua Shovan (Foundation X, 30+ years software consulting). Direct founder access during pilots; on-site visits available without travel cost.
Utah has a strong track record of healthcare and security companies that scaled nationally (Pluralsight, Domo, Qualtrics, Recursion). Healthcare systems based here have done meaningful pilots with local vendors before. We'd like to add one more name to that list.
Your next breach is a question of when, not if.
SynOI is modeled to shrink detection from months to hours, gates mass-exfiltration through human approval, and produces cryptographically signed forensic evidence your auditor can verify offline. 60-day pilot. Zero cost. Self-hosted. Your keys.