Runtime AI Assurance
for Healthcare
Monitor, control, and prove AI system behavior at every inference — across every framework and every jurisdiction.
Every AI framework is moving towards real-time governance and proof. But no single infrastructure produces it.
Building national validation infrastructure across 50+ institutional partners in 50+ countries.
AIHGle 2.0 mandates post-market surveillance of all approved AI systems.
Continuous monitoring required for all high-risk AI systems at scale.
Utah Office of AI Policy running live runtime evidence sandboxes now.
AI Airlock piloting real-world AI system testing in live clinical environments.
No standardized technical infrastructure exists to continuously prove whether an AI system still does what it was approved to do.
Three questions no single infrastructure can answer today.
What was the AI system supposed to do?
Intent baselines captured at submission. Never continuously verified against live behavior.
Did the declared safeguards actually fire?
Controls exist on paper. Independent proof of execution at each inference does not.
What changed, and when?
Models update. Populations shift. Without signed evidence, no reviewer can assess behavioral drift.
The regulatory direction is convergent. The measurement infrastructure is not.
Glacis: The Runtime AI Assurance Layer that monitors, controls, and proves.
Continuously tests AI agents, tools, retrieval, and workflows. Findings ranked by exploitability and business impact.
Installs runtime controls at the action boundary. Four control surfaces live: tool allowlist, parameter validation, approval gates, content redaction.
Signs every test, every block, every fix. OVERT 1.0 open standard. Verifiable offline by any third party.
Findings flow back into AutoRedTeam as regression tests. Loop closes automatically. The customer cannot fall behind the threat model.
Each owns part of the path: red-teaming or runtime or documentation. None ship the loop.
Find, block, sign, regress — automated end-to-end. Every stage produces evidence the next stage consumes. The loop is the product.
OVERT: An Open Evidence Standard for Runtime Assurance
OVERT is a cryptographically signed, open receipt format for runtime AI evidence. Each inference produces a compact, tamper-proof record — which controls ran, what policy applied, and what the outcome was — stored locally and verifiable offline by any third party, without exposing patient data.
Publicly versioned at overt.is. Any conformant verifier can check a receipt without Glacis involved.
Signed, chained receipts. Change one byte and the chain breaks. Verifiable offline by any third party.
Maps to EU AI Act, SOC 2, ISO 42001, or any compliance framework simultaneously. Fields and mappings are fully customizable — one evidence format serves every jurisdiction, and every organization's own standards.
Per-inference · continuous · standardized · verifiable by a third party without trusting either side
One evidence layer. Every stakeholder. Every jurisdiction.
Regulatory bodies
FDA · MHRA · HSA · EU AI Act
Does the AI system still meet requirements after deployment?
Post-market evidence at every inference
PCCP updates & adverse event support
Chain of custody no vendor can alter
Procurement & accreditation
ORCHA · NCQA · Digital Health Formularies
Do vendor compliance claims hold up in production?
Proves guardrails actually fired
Captures undisclosed model updates
Independent evidence record for listing decisions
Health systems & deployers
How do we maintain oversight across an AI portfolio?
One architecture across all AI systems
Comparable evidence across vendors
Zero PHI egress — data stays in your environment
Cross-jurisdictional operations
Standardized evidence stream across borders.
Singapore receipt verifiable in Geneva, London, Washington
Each jurisdiction keeps its own methodology
No cross-party trust required
AI vendors
How do I generate a single evidence stream that meets all the requirements I need?
One OVERT integration covers every jurisdiction
No custom compliance work per market
Evidence verifiable independently by any reviewer
From shared problem to shared infrastructure.
Institutional networks provide cross-jurisdictional reach and governance authority.
Glacis provides the runtime evidence layer and the OVERT open standard.
Scalable, adaptive, real-time AI assurance infrastructure — built to work across borders.