← AI Intelligence Hub

At the Bedside

Six Health Systems Cut EHR Time by 13+ Minutes Per Encounter Using Ambient AI Scribes

Executive Brief Six major health systems are using AI ambient scribes to give clinicians significant time back — a JAMA study confirms 13.4 fewer minutes of total EHR time per encounter, with some systems seeing documentation time drop by 27% after ten or more uses.

A study published in JAMA tracked AI ambient scribe deployment across five academic medical centers, finding total EHR time decreased by 13.4 minutes and documentation time by 16.0 minutes per encounter. Intermountain Health's Dragon Copilot deployment showed a 27% reduction in note time for clinicians who used it across 10+ encounters between April 2024 and December 2025, underscoring the compounding benefit of repeated use.

AHA Panel: AI Succeeds in Diagnostics and Radiology — But Access Equity Remains Unsolved

Executive Brief Healthcare leaders convened to assess AI's trajectory, concluding that AI is delivering in radiology and diagnostics but that smaller, under-resourced hospitals risk being left behind as larger systems accelerate deployment.

The AHA panel "AI in Health Care: Navigating Policy, Regulation, and the Road Ahead" reviewed AI's current performance across patient safety monitoring, ambient documentation, and imaging analysis. Panelists highlighted a growing access equity concern: flagship academic medical centers and large IDNs are deploying multi-modal AI stacks while critical-access and community hospitals lack both the capital and technical staff to evaluate or implement comparable solutions.

TEFCA Tops 500 Million Health Records Exchanged as HHS Leverages AI for Interoperability

Executive Brief The national health data exchange network has crossed a major threshold — 500 million records exchanged — with HHS now actively deploying AI to automate reconciliation, reduce burden, and surface insights across the interoperability infrastructure.

TEFCA, the Trusted Exchange Framework and Common Agreement, reached nearly 500 million health records exchanged across its participant network. HHS is integrating AI capabilities to lower administrative costs and reduce manual reconciliation burden, building on FHIR-native architectures that enable real-time, cross-system patient data assembly. The ONC concurrently released draft USCDI v7 on January 29, 2026, proposing 29 new standardized data elements covering adverse event reporting, nutrition, and quality improvement.

New EHR and Patient Record Integrations Bring Claude AI Directly Into Clinical Workflows

Executive Brief Anthropic's Claude is being embedded directly into EHR platforms and patient record systems, enabling clinicians to query patient history, draft notes, and surface relevant clinical guidelines without leaving their primary workflow.

Healthcare IT vendors are integrating Claude AI into EHR environments through API and MCP-based connections, allowing the model to access structured patient records, lab results, and clinical histories within enterprise security boundaries. These integrations position large language models not as standalone tools but as embedded clinical assistants that operate inside existing documentation and decision-support layers, reducing the context-switching that has historically limited AI adoption at the point of care.

From the Lab

Stanford-Harvard State of Clinical AI Report: Real-World Performance Lags Behind Controlled-Study Results

Executive Brief The first-ever joint Stanford-Harvard synthesis of clinical AI evidence draws a clear line between what AI does in a controlled study and what it actually delivers in live hospital settings — and the gap is significant enough to demand attention.

The inaugural State of Clinical AI Report 2026 synthesizes the year's most influential research, noting that while AI systems flagging deterioration risk, assisting radiology reads, and routing patient messages are now genuinely embedded in routine care, real-world deployment consistently underperforms controlled-study benchmarks. Key culprits include workflow mismatch, alert fatigue, biased training distributions, and the absence of post-deployment monitoring. The report places clinical AI in the early Slope of Enlightenment on the Gartner Hype Cycle — a healthy reckoning that separates durable value from overpromised capability.

Nature Medicine Publishes Framework for Meaningfully Evaluating AI in Clinical Settings

Executive Brief Nature Medicine's new evaluation framework addresses a persistent gap in clinical AI: most tools are validated on curated datasets under ideal conditions but never rigorously tested against what happens to real patients in real hospitals.

The paper proposes a standardized methodology for clinical AI evaluation that requires prospective validation in target deployment environments, pre-specified clinical endpoints, stratified performance analysis across demographic subgroups, and ongoing post-market surveillance tied to clinical outcomes. The framework directly responds to growing evidence that accuracy metrics derived from retrospective datasets fail to predict whether a model will improve — or inadvertently harm — patient care when deployed at scale.

Merck and Mayo Clinic Launch AI-Enabled Drug Discovery Collaboration Combining Clinical Genomics with Virtual Cell Models

Executive Brief Merck and Mayo Clinic are combining Mayo's multimodal clinical and genomic datasets with Merck's AI-powered virtual cell models in a partnership designed to accelerate target identification and early drug development — moving precision medicine from concept to pipeline candidate faster than traditional methods allow.

The agreement integrates Mayo Clinic's Platform architecture — which aggregates clinical records, genomic profiles, and proteomics data across millions of patients — with Merck's ambition to apply virtual cell simulation to disease modeling and early target prioritization. The collaboration targets oncology and immunology, where the combination of multiomics data and in silico validation has historically compressed the time from target hypothesis to lead compound by 30–50%. The partnership signals pharma's accelerating confidence that real-world clinical data combined with generative molecular models can meaningfully shorten Phase I timelines.

Deep Learning Integration of Pathology and Radiology Achieves Diagnostic Accuracy Gains in Oncology

Executive Brief A new deep learning framework that fuses pathology slide analysis with radiology imaging outperforms single-modality AI in cancer diagnosis — offering a preview of what fully integrated multimodal diagnostics will look like in routine clinical practice.

Researchers published results from the Adaptive Multi-Resolution Imaging Network (AMRI-Net), which simultaneously analyzes histopathology slides and CT/MRI scans using the Explainable Domain-Adaptive Learning (EDAL) strategy. The combined model outperformed standalone radiology and pathology AI systems on cancer staging tasks, with particular gains in cases where tissue and imaging findings diverged — precisely the ambiguous presentations that generate the most diagnostic uncertainty and missed diagnoses in clinical workflows today.

The Regulatory Landscape

WHO Europe Issues First-Ever AI in Healthcare Report: 74% of EU Member States Using AI-Assisted Diagnostics

Executive Brief A landmark WHO Europe report reveals that nearly three-quarters of EU countries have already deployed AI in clinical diagnostics — but adoption is uneven, governance frameworks lag deployment, and equity concerns are mounting across lower-income member states.

The report surveyed all EU Member States and found 74% are using AI-assisted diagnostics in some form, including tools for medical imaging, disease detection, and clinical decision support. The snapshot identifies significant variation in regulatory maturity — several member states have deployed commercial AI tools without national governance frameworks in place — and flags the lack of real-world performance monitoring as a systemic gap. The EU AI Act's August 2026 deadline for high-risk AI obligations creates urgent compliance pressure for health systems still operating under informal AI governance.

FDA Cuts Red Tape on AI Health Software and Wearables, Limiting Oversight of Low-Risk Devices

Executive Brief The FDA issued guidance in January 2026 pulling back oversight of certain AI-enabled software and consumer wearables, creating a broader deregulated zone for wellness-adjacent digital health tools — while maintaining stricter review for clinical-grade AI medical devices.

FDA's January 2026 guidance reduces scrutiny for AI-enabled software that supports general wellness functions and wearable devices that do not make specific disease claims. For clinical-grade AI medical devices, the agency maintained its transparency and labeling requirements — including mandatory disclosure that a device uses AI, documentation of model inputs and outputs, performance metrics, and identified bias sources. Most high-risk AI obligations under updated Quality Management System Regulation (QMSR, aligned to ISO 13485:2016) take effect in August 2026, with full device AI compliance required by August 2027.

Utah Becomes First State to Create a Regulatory Safe Harbor for Mental Health AI Agents

Executive Brief Utah has adopted a novel state-level approach to mental health AI regulation that reinforces consumer data protections while creating a legal safe harbor for AI agents that implement defined safety guardrails — a potential model for states navigating the vacuum left by federal inaction.

The Utah framework establishes a safe harbor for mental health AI agents that satisfy four criteria: pre-deployment safety testing, documented crisis escalation protocols, ongoing clinical oversight, and continuous post-deployment monitoring. The law simultaneously strengthens advertising restrictions and data privacy requirements for AI-based mental health tools targeting consumers. The approach directly addresses the regulatory gap exposed by the rapid proliferation of AI therapy chatbots, particularly those marketed to adolescents, without requiring full FDA medical device classification for software that falls short of clinical diagnostic claims.

Deals, Dollars & Deployments

Digital Health Funding Hits $7.4B in Q1 2026 Driven by AI Drug Discovery Mega-Rounds and M&A Rebound

Executive Brief Digital health investment surged to $7.4 billion in Q1 2026 — the strongest quarter in two years — as AI drug discovery attracted outsized rounds and M&A activity accelerated, signaling renewed investor confidence after the post-2021 correction.

Q1 2026 funding rose from $5.9B the prior quarter, with 19 mega-rounds ($100M+) accounting for 60% of all capital raised. AI's share of health tech funding has now crossed 60%, up from 55% in 2025 and 29% in 2022. The quarter's standout deal: Earendil Labs raised $787M — the largest single deal — to scale a deep learning platform that has already generated 40+ therapeutic candidates. DeepHealth's $269M acquisition of Gleamer (700+ hospital contracts) and Takeda's up-to-$1.7B collaboration with Iambic for oncology/immunology small molecules signal that consolidation is accelerating alongside early-stage investment.

UnitedHealth Group Bets $3 Billion on AI — Raising Questions About Patient Impact

Executive Brief UnitedHealth is making a $3 billion AI investment across claims processing, prior authorization, and clinical decision support — a bet that could reshape payer operations while intensifying scrutiny over whether AI accelerates coverage denials or genuinely improves patient outcomes.

The $3B commitment spans multiple AI applications including automated prior authorization review, predictive risk scoring for care management, and clinical efficiency tools for Optum's care delivery networks. STAT's reporting highlights the dual-use tension inherent in payer AI: the same predictive models that identify high-risk patients for proactive outreach can, depending on deployment, function as denial-optimization systems. UHG has not disclosed the proportion allocated to patient-facing versus administrative AI, but the scale of investment puts it among the largest single-entity AI commitments in U.S. healthcare history.

Jimini Health Raises $17M Seed Round for AI Platform Targeting Complex Behavioral Health Cases

Executive Brief Jimini Health has secured $17 million to launch Sage, an AI platform explicitly designed for complex behavioral health — going beyond the low-acuity stress and anxiety use cases that dominate the consumer mental health AI market to serve patients with serious mental illness.

Sage targets large behavioral health organizations and health systems, positioning the platform as a clinical-grade AI that operates alongside licensed clinicians rather than as a direct-to-consumer chatbot. The $17M seed represents investor conviction that the underserved complex-care segment — where workforce shortages are most acute and existing digital tools have historically failed — offers both a clinical opportunity and a defensible commercial position distinct from crowded consumer wellness apps. Jimini's approach emphasizes crisis escalation protocols and clinical oversight as foundational rather than optional features.

NVIDIA Survey: Healthcare AI Delivering Clear ROI from Radiology to Drug Discovery

Executive Brief A new survey from NVIDIA finds that healthcare organizations deploying AI across radiology, drug discovery, and clinical operations are reporting measurable returns — shifting the dominant narrative from AI as speculative investment to AI as a proven operational lever.

The survey, drawn from health system executives and life sciences leaders, found highest-reported ROI in radiology workflow automation (reduced read time, prior auth throughput), drug target identification (compressed lead generation timelines), and clinical documentation (staff time savings). Deployment barriers cited most frequently were data quality and governance gaps rather than model performance, suggesting that the bottleneck has migrated from model capability to organizational data infrastructure. NVIDIA's findings reinforce the Q1 funding data: capital is following demonstrated returns, not projected ones.

What the Field Is Talking About

Survey: Only 42% of Americans Want AI Involved in Their Care — Down from 52% in 2024

Executive Brief Public trust in healthcare AI is eroding just as clinical adoption accelerates — a widening gap between what health systems are deploying and what patients say they want that no governance framework has yet addressed head-on.

The survey found the share of Americans comfortable with AI involvement in their care dropped 10 percentage points in two years, from 52% in 2024 to 42% in 2026. The decline is sharpest among older patients and those with chronic conditions — the very populations most likely to be affected by AI-assisted triage, prior authorization, and care management tools. Healthcare AI practitioners and commentators on social media have seized on the data, with debates centering on whether transparency disclosures, explainability requirements, or patient consent mechanisms can reverse the trend before the credibility gap becomes a political liability.

'Like Drinking Salt Water': Critics Warn AI Therapy Chatbots Will Fuel Teen Mental Health Crisis

Executive Brief A pointed critique circulating widely among behavioral health professionals argues that AI therapy chatbots marketed to teenagers may worsen the very crisis they claim to address — offering the sensation of support without the therapeutic substance that actually produces recovery.

The "drinking salt water" framing — a response that mimics hydration while accelerating dehydration — has resonated with clinicians and researchers skeptical of the evidence base for consumer mental health AI. Critics argue that chatbot engagement metrics (session length, daily active users) create perverse incentives to maximize emotional dependency rather than clinical progress. The piece has generated significant commentary among psychiatrists, therapists, and health policy voices, with particular concern that at-risk adolescents may substitute AI chatbots for the evidence-based care they need but cannot access due to workforce shortages and cost barriers.