NYC Health + Hospitals is leading a controversial shift toward autonomous AI diagnostics. Explore the economic drivers, clinical risks, and regulatory battles behind this push to replace human radiologists.
On March 25, 2026, the chief executive of America’s largest public hospital system declared his readiness to begin replacing human radiologists with artificial intelligence. Mitchell H. Katz, MD, president and CEO of NYC Health + Hospitals, stated at a Crain’s New York Business panel that AI technology is now capable of handling first-pass interpretations of mammograms and X-rays autonomously. The primary remaining hurdle is regulatory approval. This aggressive push marks a critical escalation in how enterprise healthcare views artificial intelligence—transitioning the technology from a supportive clinical copilot to an autonomous primary diagnostic tool, and sparking intense debate over patient safety, liability, and the future of specialized medicine.
The Economic Imperative Driving the Shift
The driving force behind this unprecedented proposal is a combination of skyrocketing imaging demand and crippling operational costs. Radiology has traditionally been one of the most expensive and highly specialized departments in any hospital network.
Katz argued that hospitals could produce "major savings" by allowing AI to handle the initial reads of routine imaging radiologybusiness.com. Under this proposed workflow, the AI system would independently clear normal scans. Human radiologists would only be brought into the loop to provide second opinions if the AI flagged an image as abnormal.
For hospital administrators, this level of digital transformation through ai automation is no longer just about workflow optimization; it is viewed as a critical survival strategy. Sandra Scott, MD, CEO of One Brooklyn Health—a smaller hospital facing tight financial margins—supported the initiative during the same panel, noting that removing the human bottleneck for initial reads would be a "game-changer" for safety-net institutions struggling to maintain operational viability slashdot.org.
The Performance Data: Is AI Actually Better?
The readiness of hospital executives to sideline highly trained specialists stems from increasingly confident internal performance metrics. Proponents argue that for specific, high-volume tasks like breast cancer screening, AI is already outperforming human baselines.
David Lubarsky, MD, president and CEO of the Westchester Medical Center Health Network, reported that his system is already deploying advanced AI diagnostic tools with unprecedented success rates.
"The AI Westchester uses misses very few breast cancers and is actually better than human beings. For women who aren’t considered high risk, if the test comes back negative, it’s wrong only about 3 times out of 10,000."
These statistics present a compelling case for hospital boards. If an automated system can achieve a 0.03% error rate on negative screenings for low-risk populations, the financial justification for paying a human specialist to review those same negative scans rapidly evaporates.
The Danger of "AI Mirages" and Clinical Pushback
However, the medical community is raising severe alarms about the premature deployment of autonomous AI in diagnostic settings. The core concern revolves around the fundamental difference between human medical error and machine hallucination.
Recent research from Stanford University highlighted a critical flaw in current diagnostic models known as AI mirages. In these instances, AI systems successfully "ace" medical benchmark tests but do so by constructing elaborate, rational-sounding explanations for findings that do not actually exist in the X-ray marylandoutdoorclub.org.
Unlike traditional generative AI hallucinations, which are often easily identifiable by experts, AI mirages in radiology are highly coherent and logically sound, making them exceptionally difficult for oversight systems to catch. If an AI system acts as the sole gatekeeper for a "negative" scan and hallucinates a clean bill of health based on flawed visual processing, the patient is sent home without ever being evaluated by a human doctor.
This introduces complex liability questions: If an autonomous AI misses a malignant tumor, who carries the malpractice liability—the hospital, the software vendor, or the radiologist who was legally bypassed in the workflow?
Broadening Implications for Specialized Roles
The radiology debate is a bellwether for the broader tech industry and highly skilled knowledge workers. The assumption has long been that AI would automate administrative tasks and basic data entry, leaving complex, high-paying cognitive work untouched.
The NYC Health + Hospitals announcement shatters that assumption. It aligns with recent controversial statements made by Dario Amodei, CEO of Anthropic, who claimed that AI has already taken over the core functions of the radiology specialty futurism.com. While radiologists heavily disputed Amodei's claim, the fact that hospital CEOs are now actively lobbying to restructure their workforce based on this premise shows that enterprise adoption is moving faster than clinical consensus.
What to Watch Next: The Regulatory Battlefield
The immediate battleground for this transition is not technological, but legal. Current New York state regulations mandate that a licensed radiologist must read and interpret medical images.
During the March 25 forum, Katz actively rallied fellow hospital CEOs to begin pushing for changes to state regulations to allow AI to read images "without a radiologist."
Technology and healthcare analysts should closely monitor the New York State Department of Health and the state legislature in the coming months. If America's largest public hospital system successfully lobbies to rewrite the regulatory framework governing autonomous AI diagnostics, it will establish a legal precedent that will almost certainly trigger a domino effect across other states and medical specialties.
Last reviewed: April 05, 2026
