AIPODPodcast
Qure.ai runs its chest X ray algorithm with no radiologist in countries that have almost none, and as a subordinate tool where radiologists exist. The code is identical. The institutional stack is not.
In a country with two radiologists for the entire population, a routine chest X-ray taken for a cough or a pre-employment check is read by software, not a person. The patient never sees a radiologist before the algorithm flags a nodule and routes them to a CT scan that would never have been ordered otherwise. That is not a hypothetical. It is the operational reality of Qure.ai's deployment footprint across roughly 70 countries, where its chest X-ray algorithms triage tuberculosis and, increasingly, lung cancer in settings with no radiologist alternative at all. The same algorithm, in a U.S. or European hospital, is a subordinate tool. A radiologist signs every report. The clinical question is the same. The institutional structure is not.
The scale claim comes from Qure.ai co-founder and CEO Prashant Warier, who said on a June 2026 episode of the Eye on AI podcast that his company's products read about 15 million chest X-rays a year for tuberculosis screening alone, with no human in the loop in the deployments he described. Qure.ai's own homepage reports 45 million lives impacted, deployment in 105 countries, and 5,500-plus sites, numbers that predate and exceed the autonomous-TB figure. Those are company-attributed claims, and the "no human in the loop" framing is the CEO's, not an independent assessment.
The mechanism is opportunistic screening. A patient comes in for any reason that prompts a chest X-ray: a cough, a pre-operative workup, an employment physical. The same image, instead of being read only for the original indication, is also screened for tuberculosis and, on Qure.ai's lung-nodule product, for cancer risk. Patients flagged as high-risk are routed to follow-up CT or confirmatory testing. The clinical promise is that the algorithm finds disease in people who were never sent to a cancer or TB screening program in the first place.
The load-bearing clinical claim is from Qure.ai's CREATE study, which validated the company's Lung Nodule Malignancy Risk Score. Warier told the Eye on AI podcast that of 100 patients flagged as high-risk by the algorithm, 54 turned out to have malignancy on follow-up, compared with about 2 of 100 caught by standard CT screening programs. That is a roughly 27-fold enrichment, on the CEO's account. The CREATE study is company-validated, not yet independently replicated outside Qure's collaboration, and the comparison against CT screening programs is a specific framing that the underlying study design would need to support.
The "no human in the loop" line, taken literally, does not describe Qure.ai's U.S. business at all. Warier himself named the regulatory constraint: in most jurisdictions, patients cannot upload their own scans to Qure.ai's algorithms. The workflow is clinician-gated, not consumer-gated. The autonomous TB screening is real scale, and it is real in places with effectively no radiologist alternative. It is not the same product as a consumer-facing "upload your X-ray and get an AI read" service, and any reporting that conflates the two misreads the deployment geometry.
The 26 FDA clearances and 200-plus published studies that Warier cites describe the breadth of the regulatory footprint, not the depth of prospective validation for any individual product. The FDA has cleared specific Qure.ai devices for specific clinical claims; that is regulatory permission, not blanket diagnostic approval. The forward claim, that primary care will be "AI-first" within five to ten years, is a CEO prediction, not a finding, and is best read against the bottleneck Warier himself identified: who gets to decide what the algorithm reads, and who is liable when it is wrong.
The question for the next decade of primary care is no longer whether AI can read a chest X-ray. The technical answer is already yes, at scale, in places where the alternative is no read at all. The open question is regulatory, and it is the one the reader has standing in: should a patient in any country be able to upload their own scan and receive an AI read directly, and who is liable when the read is wrong. Warier's 15-million figure is the answer to the first question. The 26 FDA clearances and the upload-your-own-scan bottleneck are the answer to the second. They are different questions, and the deployment footprint Qure.ai describes lives in the gap between them.