Clinical & Diagnostics
Mount Sinai Integrates OpenEvidence AI into Epic EHR System-Wide — Nurses and Pharmacists Included
OpenEvidence functions as a governed medical search engine that accepts natural-language clinical questions and returns answers strictly grounded in peer-reviewed literature and clinical practice guidelines — by design reducing hallucination risk. The Epic integration solves the last-mile adoption problem: clinicians query the AI without leaving their existing workflow. This marks OpenEvidence's first enterprise-scale deal that explicitly covers the full clinical care team rather than restricting access to attending physicians.
JAMA Study: AI Scribes Cut Physician Burnout by 13.9 Points — But Time Savings Are More Modest Than Advertised
Across 186 clinicians and over 1,800 adopters at five major health systems, ambient AI scribes delivered a net 13.9-percentage-point reduction in burnout and a 6.2-point reduction in severe burnout. EHR time savings measured 16 minutes in documentation and 13.4 minutes in total EHR time per 8-hour shift — meaningful, but below what some vendor claims suggest. The study also found improvements in after-hours charting time, cognitive task load, and patient face-time. Researchers note the psychological benefits may outpace the raw time gains.
Aidoc Wins FDA Clearance for Healthcare's First Comprehensive Foundation Model AI — 14 Acute Conditions, One Model
The clearance covers Aidoc's CARE™ foundation model, which covers 11 newly cleared indications — including appendicitis, bowel obstruction, kidney stone, intestinal ischemia, and pelvic fracture — plus three prior clearances, all running on a single unified model. In the FDA-reviewed pivotal study, the 11 new indications achieved mean sensitivity of 97% (up to 98.5%) and mean specificity of 98% (up to 99.7%). The supporting aiOS platform has now analyzed over 100 million patient cases. This represents the first FDA clearance of double-digit acute indications powered by one foundation model.
Research & Science
Stanford–Harvard ARISE Network Releases "State of Clinical AI 2026" — AI Works, but Brittleness and Over-Reliance Persist
The report synthesizes the most influential clinical AI studies published in 2025. Strongest results appear in prediction tasks — early-warning deterioration models, risk scoring, disease trajectory forecasting — and in AI-assisted radiology where physicians used AI as an optional second opinion. Patient-facing AI expanded rapidly but lacks rigorous outcome-based evaluation; most studies measured engagement, not health outcomes. A key safety finding: some studies documented clinicians following incorrect model recommendations even when errors were detectable, flagging the over-reliance problem as under-examined in deployment research.
NVIDIA Survey: 70% of Healthcare Firms Now Deploy AI — 85% Report Revenue Gains as ROI Shifts from Aspiration to Reality
The survey of healthcare and life sciences executives shows adoption at 70% (up from 63% in 2024). 85% of respondents report AI helping increase revenue; 80% report cost reductions. 44% of management respondents say AI has boosted annual revenue by more than 10%. By sector: 57% of medical technology companies cite measurable imaging AI ROI; 46% of pharma/biotech firms cite drug discovery as a top ROI driver. Clinical decision support, medical imaging analysis, and workflow optimization rank as the top three deployed use cases. 85% plan to increase AI spending in 2026.
Deep Learning Framework Merges Pathology and Radiology in a Single Diagnostic AI System — Nature Scientific Reports
The model applies transformer-based architectures to simultaneously process imaging (radiology) and tissue (pathology) data, enabling joint representation learning across modalities. The framework addresses cross-cohort robustness — a persistent failure point for single-modality models — through shared embedding spaces trained on paired radiology-pathology datasets. Researchers highlight applications in oncology, where combining scan and biopsy data yields improved diagnostic accuracy and prognostic predictions beyond what either modality provides alone. The work builds on the broader radiomics trend of extracting quantitative patterns invisible to human readers.
Policy & Regulation
FDA Announces Sweeping Pullback on Oversight of AI-Enabled Software and Consumer Wearables
Under the new guidance, clinical decision support software — including AI and generative AI features — escapes FDA device regulation if it delivers a single, clinically appropriate recommendation and allows the clinician to independently review the underlying logic and data. Wearables providing heart rate, blood pressure, and blood glucose readings for wellness purposes gain expanded latitude. High-risk diagnostic and therapeutic AI products remain under full device oversight. The CPT 2026 code set simultaneously added 288 new codes covering digital health and AI services, signaling CMS alignment with wider AI reimbursement. Critics note the guidance leaves a regulatory vacuum for mid-risk AI tools.
Georgia Passes Law Banning AI-Only Insurance Coverage Decisions — As Federal AI Legislation Remains Fragmented
The Georgia statute targets automated prior-authorization and coverage-denial workflows increasingly powered by large language models and clinical AI systems. It mandates human review at the final decision layer — a safeguard against systems like the AI-driven utilization management tools that multiple class action lawsuits have targeted over the past two years. The law does not ban AI from informing recommendations, only from making final binding coverage decisions autonomously. With no federal equivalent in place, payers operating across state lines now face a patchwork compliance environment as similar bills advance in other state legislatures.
HHS Announces TEFCA Milestone: Nearly 500 Million Health Records Exchanged Across America's National Interoperability Network
TEFCA uses FHIR-native architecture and a network of Qualified Health Information Networks (QHINs) to enable bidirectional data sharing across participating providers, payers, and public health agencies. HHS is leveraging the network's scale alongside AI tooling to reduce administrative burden and lower costs. ASTP/ONC simultaneously released draft USCDI v7 in January 2026, proposing 29 new data elements to strengthen nationwide interoperability. For AI developers, TEFCA's growth means larger, more representative training datasets and more reliable real-time clinical data feeds for deployed models.
Industry & Business
Qualified Health Raises $125M Series B to Scale Enterprise AI Across U.S. Health Systems — Anthropic Among New Investors
Qualified Health now supports 400,000 users representing ~5% of U.S. hospital revenue. The platform builds secure data foundations across EHR and non-EHR sources, deploys assistants and automated workflows, and includes governance tooling for tracked AI deployment. At UTMB, the company generated over $15 million in measurable run-rate impact within six months of deployment. New partnerships with Jefferson Health (Philadelphia) and the University of Texas System anchor the expansion, with use cases spanning quality registry solutions, care gap surfacing, and administrative task automation. The round was led by NEA.
Jimini Health Raises $17M for Sage, a Clinician-Supervised AI That Treats Mental Health Patients Around the Clock
Sage's underlying model is fine-tuned on clinician-authored care plans for each patient and operates within strict behavior guardrails that prevent deviation from those plans. Jimini runs its own clinical practice employing licensed clinicians who treat real patients before any model update is deployed at scale — a live red-teaming approach unusual in healthtech. The seed round was led by M13 and Zetta Venture Partners, with Town Hall Ventures, LionBird, and OneMind participating. CEO Luis Voloch previously founded Immunai (valued at $1B+). The backdrop: STAT News reports that more than 1 million people per week have ChatGPT conversations showing indicators of suicidal planning or intent — underscoring why the supervised-AI-only model is commercially differentiated.
Translucent Raises $27M Series A Led by Google Ventures to Give Every Hospital Its Own AI Financial Leader
The platform covers six financial domains simultaneously: claims, labor, clinical output, P&L, budgets and forecasts, and contract economics. It uses agentic AI to automate root cause analysis and variance identification — work previously done manually by finance analysts. Early customers including Northwestern Medicine and Springfield Clinic report completing 97% of routine financial analysis without manual effort and a 56% increase in finance team capacity without adding headcount. GV led the Series A alongside NEA, Virtue, and FPV Ventures. The company was founded in 2024 by Jack O'Hara.
AstraZeneca Acquires Modella AI to Embed Multimodal Foundation Models into Its Global Oncology R&D Pipeline
Modella AI's multimodal foundation models and AI agents will be embedded directly into AstraZeneca's oncology R&D organization to accelerate clinical development, enhance biomarker discovery, and automate data-intensive workflows across the drug pipeline. The technology integrates pathology, genomics, and imaging data streams for joint analysis — enabling generation of biological insights across AstraZeneca's global portfolio at scale. Financial terms were not disclosed. The deal expands a partnership initiated in July 2025 and was announced at JPM26 alongside AstraZeneca's broader AI-in-R&D strategy.
Roche and NVIDIA Launch AI Factory to Accelerate Development of Therapeutics and Diagnostics at Scale
The AI factory model — popularized by NVIDIA as a framework for industrial-scale AI production — applies high-performance GPU clusters to continuously generate, evaluate, and deploy AI models for biomedical workflows. For Roche, this enables training on large proprietary multi-omics and imaging datasets that would be cost-prohibitive on standard infrastructure. The collaboration spans diagnostics (pathology AI, biomarker identification) and therapeutics R&D (target discovery, clinical trial optimization). This follows NVIDIA's broader push into healthcare AI infrastructure, now spanning partnerships with major health systems, pharma companies, and device manufacturers.
Social Buzz
MedCity News Op-Ed: "Healthcare's AI Obsession Is Missing the Point on Nursing Shortages" — And the Comments Are Fierce
The piece engages directly with the actual data: the U.S. healthcare sector faced shortages of 250,710 RNs in 2025, with 65%+ of hospitals operating below full capacity due to staffing gaps. The author's argument cuts against the prevailing vendor narrative — that AI scheduling tools, predictive burnout models, and medication delivery robots will close the gap — contending that these tools optimize around a broken system rather than repairing it. The response from nursing leaders on LinkedIn and X was extensive, with threads from frontline nurses describing the experience of AI tools that add documentation overhead while headcount cuts continue. The debate reflects a broader tension in 2026 between operational AI optimism and clinical workforce skepticism.
Stanford Medicine's X Post on the State of Clinical AI Report Goes Viral — Sparks Debate on AI Safety and Over-Reliance
The viral post linked to the ARISE network report covering 2025's most influential clinical AI literature. The over-reliance finding generated the most discussion: studies documented cases where clinicians deferred to model outputs in the presence of detectable errors — a phenomenon researchers tie to alert fatigue, institutional pressure to use AI tools, and inadequate training on model limitations. LinkedIn posts from clinical informaticists noted the gap between vendor performance metrics (sensitivity, specificity on validation sets) and real-world deployment behavior. The thread reinforced the emerging consensus that clinical AI governance — not just model accuracy — is the field's most urgent unsolved problem in 2026.