Standards define what trustworthy AI should do.
We test what it actually does.
Gate Zero is the checkpoint between a vendor’s sandbox and a hospital’s live system. Standardized, high-volume, extreme stress-testing performed outside the vendor and outside the hospital — before a system is trusted with a patient, and after, for as long as it stays in use.
Real progress. None of it independently tested.
NIST, CHAI, the Joint Commission, and ONC/ASTP have each built real, substantive pieces of what oversight should look like — faster than anyone could reasonably have expected. What none of them can do is manufacture, inside every hospital, the skill set required to check a live clinical model against them.
Defines the risk-management functions a trustworthy AI program should have in place.
A well-built, genuinely useful baseline for documenting a tool’s intended use and limitations.
Expects dated, current artifacts — bias testing, pre-deployment validation, production drift monitoring.
Requires a change-management plan and vendor disclosure of validation and monitoring information.
“Owning the software and running the software weren’t proof of anything.”
— FROM “THE PILOT PROBLEM,” THE LAB’S FOUNDING ESSAYStandardized, high-volume stress-testing across every surface a clinical AI touches.
The same premise as Leapfrog’s CPOE Evaluation Tool, rebuilt for agentic and workflow-integrated healthcare AI: a defined battery of scenarios, run by an outside party, against the live system.
Synthetic FHIR / EHR
Full synthetic patient records and chart histories, built to exercise real documentation patterns.
HL7 & device interfaces
Connected monitors, orders, and results traffic at realistic volume and timing.
Voice & ambient capture
Ambient documentation and voice-driven workflows under real clinical noise conditions.
Nurse-call & workflow systems
Escalation paths and alerting behavior across the systems a response actually depends on.
Standardized patients
Scripted patient scenarios built the way clinical training already validates a human clinician.
Hidden scenario libraries
Test cases the vendor doesn’t see in advance — probes, edge cases, and known failure patterns.
Failure injection
Deliberately degraded data, dropped signals, and conflicting inputs to find where behavior breaks.
Clinician panel review
Working clinicians adjudicate what “wrong” actually looks like in context — not just what’s statistically off.
One gate. Before go-live, and for as long as the system is in use.
Baseline & Stress
Independent pre-deployment validation against the hidden scenario library, sized to the system’s intended use.
Gate Zero
A dated, current disposition record — findings, coverage, human review — a hospital can actually show a surveyor. Nothing passes through ungated.
Monitor
Continuous or periodic re-testing as the model updates, drifts, or meets patients a pilot never covered.
Findings are the product, not the marketing.
Aggregate, anonymized results are published on a set schedule — independent of any single vendor relationship.
The Pilot Problem
Position essay — what CPOE and the Leapfrog Evaluation Tool already taught healthcare about independently checking safety technology.
The Missing Gate
Companion methods paper proposing the Lane 0 (Sandbox) → Lane 1 (Build) evidence gate for rapidly developed healthcare software.
First-Test Report
Results from the lab’s first full validation engagement, tied to pilot data.
Aggregate Findings
Recurring, anonymized cross-engagement reporting once cross-product data exists.
Jennifer Foster
A health informatics and clinical systems interoperability leader with 25 years in enterprise healthcare technology and the regulatory reporting that satisfies CMS, Joint Commission, and health-plan requirements — grounded in a BSN, RN foundation. She has led enterprise governance programs, participates in AI governance work, including a seat on CHAI committees, and has spent her career translating clinical and technical requirements for CEOs and boards. Rather than writing interoperability standards, her work ensures hospitals and vendors actually follow them — verifying compliance, then owning the testing and implementation that puts it into practice. At Gate Zero, she runs operations and handles the details — the same rigor that, for a quarter century, has meant hospitals could trust that their systems did what they were supposed to.
jenfoster@gatezeroai.com
Dr. Joseph Izzo
An emergency and addiction medicine physician with an MD from Rutgers Robert Wood Johnson Medical School and a chemistry, computer science, and mathematics background from NYU, Dr. Izzo has spent his career at the exact intersection Gate Zero exists to guard: the point where clinical judgment, live data, and technology collide under pressure. As a Chief Medical Information Officer, he has led digital innovation and AI governance programs, and holds the unusual combination of board certification in Clinical Informatics and Addiction Medicine (ABPM) alongside CompTIA Security+ and PenTest+ — training built for finding what a system misses before it costs someone their life. He is also CHIME-credentialed (CDH-E), PMP-certified, and a Diplomate of the American Board of Emergency Medicine. At Gate Zero, he owns product, EHR and test-harness architecture, security, and the testing methodology behind every evaluation the lab runs.
josephizzo@gatezeroai.comNina Izzo
Advises on legal structure and regulatory positioning across the NIST, CHAI, FDA, and ONC/ASTP landscape.
Nothing reaches the hospital without going through the gate.
Gate Zero is currently in founding-stage development. Early conversations with health systems, vendors, and advisors are open now.