The head of the largest public health system in the United States has suggested that artificial intelligence could eventually take over much of the work now done by radiologists, a claim that collides directly with New York state regulations requiring board-certified physicians to interpret medical images. The assertion, attributed to the CEO of NYC Health + Hospitals, has drawn sharp attention because it arrives alongside a wave of new AI legislation in New York and fresh clinical evidence from Europe showing that algorithms can dramatically cut radiologist workloads in cancer screening. Whether that vision can survive the state’s existing legal framework is a separate question entirely.
What is verified so far
The strongest piece of clinical evidence supporting the idea that AI can shoulder routine radiology work comes from a prospective paired noninferiority trial conducted in Spain. Published in Nature Medicine, the study screened women from March 15, 2022, to January 11, 2024, using AI-based triage and decision support in mammography and digital tomosynthesis for breast cancer screening. Low-risk cases were automatically classified as normal by the algorithm, and the trial reported a greater than 60% workload reduction for radiologists without compromising detection accuracy. That figure is significant: it suggests that for routine breast cancer screening, the majority of reads could be handled by software rather than a human specialist.
The Spanish trial’s design matters. As a prospective, paired noninferiority study, each mammogram was assessed both with and without AI support, allowing investigators to compare performance directly. According to the published protocol, the primary outcome was cancer detection rate, with secondary outcomes including recall rates and reading time. Across these measures, AI-assisted workflows maintained diagnostic quality while substantially reducing the number of images requiring full radiologist review. The study did not eliminate radiologists; instead, it repositioned them as supervisors of edge cases and complex findings.
On the regulatory side, New York has been building a layered AI governance structure. Governor Kathy Hochul signed the RAISE Act, a measure described in a state announcement as requiring transparency and reporting from developers of large frontier AI models. The law mandates 72-hour incident reporting for significant AI-related harms and creates an oversight office within the New York Department of Financial Services. Its focus is broad, covering sectors such as finance and critical infrastructure, but it signals that state officials are preparing to scrutinize powerful AI systems across multiple domains.
Separately, state lawmakers have introduced Senate Bill 2025-S7263, which would impose liability for damages caused by a chatbot impersonating certain licensed professionals. While the bill is not limited to medicine, its language clearly contemplates situations in which AI tools could mislead the public by mimicking doctors, lawyers, or other credentialed experts. If enacted, it would create a new layer of legal risk for healthcare organizations that deploy conversational AI in patient-facing roles without clear safeguards.
Yet the legal barrier to replacing radiologists is not hypothetical. Under N.Y. Comp. Codes R. & Regs. Tit. 10 Section 405.15, hospitals must ensure that “radiologic tests shall be interpreted by a board certified or board admissible radiologist,” with only limited exceptions. That rule, as summarized by Cornell Law, means that no matter how accurate an AI system proves to be, a licensed human must still sign off on imaging results in New York hospitals. Any plan to let software replace radiologists would require either a regulatory amendment or a creative legal interpretation that current language does not obviously support.
New York’s broader institutional landscape also shapes how quickly AI could move into clinical workflows. The state’s official portal at ny.gov highlights a growing number of digital initiatives, but it does not yet spell out a comprehensive policy for AI in direct patient care. Agencies such as the Office of Language Access, which oversees translation and interpretation services through the state language access office, illustrate how New York tends to build specialized oversight structures when communication or comprehension is at stake. A similar, dedicated framework for clinical AI has not been publicly articulated.
What remains uncertain
The most important gap in this story is the absence of a verified primary statement or transcript from the NYC Health + Hospitals CEO. Secondary news reports have attributed the claim about AI replacing radiologists to the system’s leadership, but no official press release, internal memo, or recorded speech transcript from NYC Health + Hospitals has been made publicly available to confirm the exact wording, context, or scope of those remarks. Without that primary record, it is difficult to assess whether the CEO was describing a near-term operational plan, a long-range aspiration, or simply commenting on the general direction of the technology.
Equally unclear is whether NYC Health + Hospitals has begun any formal AI integration in its radiology departments. No policy document or pilot program announcement has surfaced. The system serves a large patient population across New York City’s public hospitals, and any shift in how imaging is handled would carry real consequences for both staff and patients. But the available evidence does not confirm that such a shift is underway or imminent, nor does it show that the system has sought a waiver or reinterpretation of the state’s radiology rules.
There is also no updated guidance from the New York Department of Health on how AI-assisted interpretations interact with the existing requirement under Tit. 10 Section 405.15. The regulation was written for a world in which only human physicians read imaging results. Whether the state views AI triage of low-risk cases as a permissible preliminary step, or as an outright violation when performed without immediate physician oversight, has not been clarified in any public regulatory bulletin or advisory. The RAISE Act addresses frontier model developers broadly, but its oversight mechanism sits within the Department of Financial Services, not the Department of Health, leaving open questions about how healthcare-specific AI applications will be supervised.
The Spanish mammography trial, while rigorous in design, studied a population and clinical setting that may not translate directly to the diverse patient demographics and operational realities of a large American public hospital system. Screening programs in Spain differ in participation rates, prior imaging availability, and follow-up infrastructure. Peer-reviewed evidence from a single European cohort carries weight, but it does not settle the question of whether AI can safely replace radiologists across all imaging modalities, from CT scans to MRIs, or in populations with higher comorbidity burdens and more heterogeneous imaging equipment.
Finally, the labor and ethical dimensions remain largely uncharted. Radiologists in New York operate under a clear legal mandate but face growing pressure to handle rising imaging volumes. Whether professional societies, unions, and hospital administrators would accept a model in which AI handles most normal studies while humans focus on complex cases is still unknown. Without clear statements from these stakeholders, predictions about workforce displacement or role redefinition remain speculative.
How to read the evidence
The available evidence falls into three distinct categories, and readers should weigh each differently. The strongest material is the clinical trial data and the regulatory text. The Spanish mammography study is a prospective, paired noninferiority trial published in a top-tier medical journal, which places it among the most reliable forms of clinical evidence. Its finding of greater than 60% workload reduction is specific, quantified, and methodologically transparent. The New York regulatory code requiring board-certified radiologist interpretation is black-letter law, not currently softened by exceptions for AI. These two facts form the hard floor and ceiling of the debate. AI can demonstrably do much of the screening work, but New York law presently forbids hospitals from letting it do so without physician sign-off.
The legislative developments occupy a middle tier of evidence. The RAISE Act is signed law, and its provisions for 72-hour incident reporting and a new oversight office are concrete and enforceable, though they address AI system providers more directly than hospital users. Senate Bill S7263 is still pending, which means its liability provisions for chatbot impersonation of licensed professionals are hypothetical rather than operative. Still, both measures show that lawmakers are already thinking about how AI might blur the line between licensed expertise and automated advice, a concern that resonates with the idea of algorithms reading medical images.
The weakest tier is the unverified attribution of forward-looking comments to the NYC Health + Hospitals CEO. Without a primary source, those remarks cannot be treated as a firm policy statement or a detailed roadmap. At most, they suggest that senior leaders in a major public health system are paying attention to AI’s potential in radiology and are willing to discuss scenarios in which software takes on a larger share of the work. How that interest will interact with entrenched legal requirements, patient safety expectations, and professional norms in New York remains an open question.
For now, the most grounded conclusion is narrow but important: AI has shown that it can substantially reduce radiologist workloads in at least one well-studied screening context, yet New York’s current regulatory framework still requires human radiologists to interpret hospital imaging studies. Any future in which algorithms replace, rather than assist, those specialists would demand not only more clinical evidence across diverse settings, but also explicit changes to state law and clear guidance from health authorities. Until then, AI in New York radiology is best understood as a powerful adjunct tool constrained by rules that were written for a different technological era.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
