Clinical AI tools that are ready for widespread use are not reaching most hospitals. The reason is financial, not clinical. HHS has fixed this exact problem twice before, for electronic health records and for cybersecurity, by letting industry fund adoption under strict guardrails while a permanent payment system caught up. It should do the same for AI: a new safe harbor letting life sciences and technology companies fund adoption of AI clinical pathway tools that are product neutral, grounded in peer-reviewed evidence or established clinical guidelines, and free of sponsor influence over clinical output.
An AI tool that flags a stroke on CT and coordinates the response across every provider on a case is now standard practice at many hospitals, because it secured Medicare reimbursement early. In a study of 166 hospitals and more than 14,000 patients, it cut time to treatment by nearly 40 minutes, a 44% reduction. Compare that to hypertrophic cardiomyopathy, a genetic heart condition that can cause sudden cardiac arrest in people who appear healthy. AI tools can now flag suspect patients on a routine EKG years before symptoms appear, but roughly 60% of cases are still misdiagnosed, often after four incorrect diagnoses. The clinical logic is the same in both cases. What differs is whether a reimbursement code exists. Not surprisingly, health systems consistently cite the lack of reimbursement as a primary barrier to adopting more AI tools.
CMS is acting. It has proposed a new payment category, Software as a Medical Service, to standardize reimbursement for AI tools that read a scan, an EKG, or a chart and return a diagnosis or risk score. It has stood up a dedicated AI office and is testing outcome-based payment models for technology-enabled chronic care. The first wave alone targets 36 new codes. But all of this moves at the pace of rulemaking and claims-data cycles, measured in years, while AI tools keep emerging faster than that. Once a code exists, adoption tends to follow. The years before a code exists are where an already validated tools sits unused.
CMS built a bridge for exactly this gap two decades ago when it launched the New Technology Add-on Payment (NTAP) program to provide temporary reimbursement for a handful of new drugs and devices each year while a permanent pathway was established. NTAP was never sized for the current volume of AI tools clearing FDA review, and CMS has proposed retiring it. That would remove one of the few bridges hospitals have during the waiting period, just as the number of tools waiting behind it is growing fastest.
Federal regulators have solved a version of this problem twice before. Since 2006, HHS has protected the donation of electronic health record software under a dedicated safe harbor and Stark Law exception, on the view that getting infrastructure into hospitals faster served patients. In 2020, facing a similar gap in cybersecurity, OIG did it again for donated cybersecurity technology. Hospitals now run donated EHR and cybersecurity infrastructure at scale, and neither arrangement is regarded as problematic under the Anti-Kickback Statute.
The same tool, applied to AI clinical pathway technology, closes the current gap without waiting for CMS rulemaking to catch up. Life sciences companies already fund clinical trials inside hospitals under comparable guardrails: independent clinical judgment, no product promotion, transparent evidence standards. An AI Clinical Pathway Safe Harbor would permit life sciences and technology companies to fund adoption of AI tools that meet three conditions: the can’t favor one company’s drug, device, or service over another’s; it has to be grounded in peer-reviewed evidence or established clinical guidelines; and the sponsor can’t have a hand in what the tool actually tells the clinician. In plain terms: a life sciences company can pay for the infrastructure, but it can’t buy influence over what a doctor is told to prescribe or order. This fits comfortably inside logic OIG has already applied twice.
The cost of waiting is not evenly distributed. Rural and critical access hospitals report under 50 percent adoption of even basic predictive tools, compared with 86 percent among top-tier system-affiliated hospitals. Every year without a bridge widens that gap, and it widens it fastest for the hospitals with the least room to absorb it.
HHS has the authority to act today, through the same rulemaking process it used for the EHR and cybersecurity safe harbors: propose a rule, take comment, finalize it. The clinical evidence for these tools is not in question. What is missing is the financing bridge to get them into the hospitals that need them most, and HHS has already built that bridge twice.
Photo: Getty Images
Andrew M. Ibrahim MD, MSc is a practicing surgeon at the University of Michigan and the Chief Clinical Officer at Viz.ai, an agentic AI platform deployed across 2,000 hospitals. Viz.ai received the first FDA de novo clearance for AI-aided triage and notification and was the first AI technology to receive a New Technology Add-On Payment (NTAP) from CMS.
This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.
