AINEWS
Three Israeli medical AI systems that perform defined clinical tasks, including at home pregnancy ultrasound and heart failure triage, are now under live regulator oversight to ask: who is on the hook when the algorithm makes the call.
Israel's regulators have launched a test bed for medical AI that does more than flag a likely diagnosis. The Israel Innovation Authority and the Israel Ministry of Health's jointly announced sandbox puts three Israeli systems that perform defined clinical tasks under live clinical oversight for the first time. The question the program is built to answer is what happens when the algorithm, not the physician, is the one making the call.
The first cohort, announced in the Digital Journal analysis, is small. The three companies are clinical cases, chosen because each one pushes the line between assistive and autonomous in a different setting.
Pulsenmore will pilot an at-home prenatal ultrasound system with Beilinson Hospital, part of the Rabin Medical Center network inside Clalit Health Services. The company says its device has been used for more than 250,000 home scans, and the IIA is funding a $1 million project for the AI component, both figures drawn from Pulsenmore's own commercial and grant disclosures. The company separately holds FDA De Novo authorization for the home-use device in the United States, a separate regulatory track that puts the Israeli pilot in a different evidentiary position than a greenfield approval.
Cordio Medical will run a heart-failure management system at Tel Aviv Sourasky Medical Center (Ichilov Hospital) that combines voice analysis, patient questionnaires, smartphone and smartwatch data, and the hospital's electronic medical record. The system is designed to detect deterioration and recommend medication adjustments under a physician-defined protocol, not to prescribe on its own. The qualifier is the point. This is decision support at the edge of autonomy, not autonomous prescribing.
The third company, rendered "Simahook" in the Digital Journal analysis and "Simhawk" in the JNS wire carried by Cleveland Jewish News, will run AI-guided fetal weight assessment trials at Hadassah Ein Kerem and Hadassah Mount Scopus. The transliteration will likely resolve once the IIA publishes the official roster; for now the program is the story, not the spelling.
Health Ministry spokesperson Shira Solomon announced the cohort. Director-General Moshe Bar Siman Tov and Israel Innovation Authority chairman Alon Stoppel framed the program as a way to learn regulation alongside industry rather than write it in advance. The framing fits a category of medical AI that existing device frameworks were not built to govern: systems that act on the patient within a defined clinical protocol, rather than systems that read a scan and leave the decision to a human.
The UK's MHRA has launched its own AI sandbox, but the British program is medicines-development-focused rather than healthcare-AI-specific. The EU AI Act, in force since 2024, takes a risk-tier approach to medical-device AI but does not operate a clinical test bed. Health Canada's pre-market guidance for machine-learning-enabled medical devices, including its Predetermined Change Control Plan, is closer in spirit to what Israel is attempting but stops at the pre-market line. The World Health Organization's ethics-and-governance guidance sets the international floor. None of these is a clinical sandbox for systems that act.
That gap is the contribution Israel is offering to test. A sandbox is not a relaxation of standards. It is a controlled environment in which the standards have to be invented, evidence by evidence, before any of these systems ship at scale. The harder questions, on how much human oversight is enough, what evidence proves safety across patient groups and care settings, and who is accountable when an algorithm acts, sit exactly where the technology stops assisting and starts acting.
The first results from the Israeli cohort will not resolve those questions. They will define them.