For most of the last decade, AI in radiology has meant one algorithm for one problem: a model that flags a pulmonary embolism, another that flags a brain bleed, another for a spine fracture. On January 21, 2026, Aidoc said the U.S. Food and Drug Administration cleared something different — a single foundation model, branded CARE, that can flag 14 acute findings on an abdominal CT scan at once, folding 11 newly cleared indications together with three the company already had into one workflow. It is being described inside the company as healthcare’s first comprehensive foundation-model AI triage solution, and it is already running in more places than almost any rival clinical AI product on the market.
One Scan, Fourteen Possible Emergencies
Abdominal CT is one of the most heavily ordered scans in any emergency department, and it can hide dozens of unrelated emergencies in a single image — a bowel obstruction here, a kidney stone there, a hemorrhage somewhere else entirely. Historically, catching all of those meant running a stack of narrow, single-purpose algorithms over the same image, each with its own false-positive rate, its own alert, and its own chance of being ignored amid alarm fatigue. Aidoc’s CARE model was built to scan for all 14 conditions in one pass, feeding results into the company’s aiOS enterprise operating system rather than a pile of disconnected pop-ups.
The Numbers Behind the Clearance
Aidoc says the cleared model posted a mean sensitivity of 97%, reaching as high as 98.5% depending on the finding, alongside a mean specificity of 98%, topping out at 99.7%. Just as notable is the company’s claim of roughly a tenfold reduction in false alerts compared with running separate single-condition tools side by side — a figure aimed squarely at the complaint radiologists raise most often about AI triage software: that it cries wolf so often clinicians start tuning it out. Aidoc is already the most widely deployed pure-play radiology AI vendor in the world, with tools running in more than 1,600 medical centers and processing over 60 million patient scans a year, giving the company an unusually large real-world base to point to.
What Hospitals Say They’re Actually Getting
Dr. Heidi Beilis, chief medical officer at WellSpan Health, said pulling key acute conditions into a single workflow represents “a fundamental shift in how radiology departments operate,” language that captures what health systems are really buying: not just detection accuracy, but a way to cut through the noise of emergency department crowding and imaging backlogs that have become chronic problems at U.S. hospitals. Aidoc CEO Elad Walach framed the clearance as a deliberate trade-off the company made in building CARE, saying that “breadth alone wasn’t enough” and that the team was “committed to meeting the safety and quality threshold required for real-world clinical use” — an acknowledgment that regulators and hospitals have grown warier of AI tools that promise broad capability without matching evidence.
Skeptics Still Want to See It at Scale
Radiology AI has a mixed track record of translating strong validation numbers into durable clinical value once tools leave controlled study conditions and meet messy real-world caseloads, inconsistent scanner hardware, and clinician workflows that vary hospital to hospital. Critics of foundation-model approaches in particular note that a single model handling 14 conditions concentrates risk: if the underlying model has a blind spot, it could affect every one of those findings rather than just one narrow use case, unlike a fragmented ecosystem of separate single-purpose tools. Proponents counter that consolidating detection into one validated model, subject to one clearance and one monitoring pathway, is actually easier to audit and update than the alternative of dozens of siloed vendors and algorithms running in parallel with no shared oversight.
A Bigger Bet on Foundation Models in Medicine
Aidoc’s move lands amid a broader industry shift away from narrow, single-task AI tools and toward foundation models that can be pointed at many problems within one imaging modality. It also arrives as the FDA itself has been rethinking how it oversees AI-enabled devices, moving toward predetermined change control plans that let cleared models update over time rather than forcing a new submission for every tweak — a regulatory posture that foundation-model vendors like Aidoc are counting on to keep pace with fast-moving software.
What to Watch Next
The real test for CARE will not be the clearance itself but adoption data over the next year: how many of Aidoc’s 1,600-plus hospital sites actually turn the comprehensive model on, whether the promised drop in false alerts holds up across different patient populations and scanner brands, and whether rival radiology AI vendors respond by rushing their own multi-condition foundation models through the FDA pipeline. If the tenfold reduction in false alerts proves durable, it could reshape how emergency radiology workflows are built industry-wide; if it doesn’t, CARE will join a long list of AI tools that looked transformative on a spec sheet and modest at the bedside.
