← AI Intelligence Hub

Clinical & Diagnostics

Hospitals Roll Out Chatbots, Looking to Reclaim Their Role in Patients' Health Conversations

Executive BriefHospitals are deploying their own AI patient chatbots — before patients turn to Google, ChatGPT, or Reddit — positioning health system-owned AI as the first stop for health questions. Hartford HealthCare launched PatientGPT and two major systems are now piloting Epic's Emmie, marking a strategic shift in who owns the patient AI relationship.

Hartford HealthCare launched PatientGPT, built on K Health's clinical AI platform, to patients across Connecticut. Sutter Health and Reid Health (Indiana/Ohio) are piloting Epic's "Ask Emmie," a conversational assistant embedded in MyChart that grounds responses in each patient's longitudinal medical record — not generic medical content. Emmie answers questions about lab results, medications, and care plans within MyChart's credentialed environment, with complex queries routed to human clinicians. The architecture keeps responses anchored to validated patient data rather than probabilistic generation alone.

Sutter Health Becomes First Health System to Go Live With Epic's AI Patient Chatbot

Executive BriefSutter Health is the world's first organization live with Epic's Ask Emmie — a patient-facing AI assistant that answers health questions grounded in the individual's own medical record, not general internet content. This is the opening move in what will likely become a standard feature across Epic's 2,000+ health system clients.

Ask Emmie integrates directly into the MyChart patient portal, using each user's longitudinal medical record as the retrieval context for language model responses. The tool answers questions about recent test results, appointment instructions, and medication queries within a clinically validated framework. Unlike general-purpose AI, Emmie's responses are bounded by the structured patient record, reducing the risk of hallucination on clinically critical questions. Sutter Health worked with Epic to become the first organization in production, with Reid Health the second pilot site.

Clinical Trials in 2026: Platformization, AI Fluency, and the Redrawing of the Value Chain

Executive BriefAI is no longer a bolt-on tool in clinical research — it is reshaping the fundamental architecture of how trials are designed, enrolled, and analyzed. Organizations that treat AI fluency as a core competency are compressing trial timelines and surfacing better evidence; those that don't are being outcompeted.

AI-powered simulation platforms now model patient population characteristics, predict dropout rates, and stress-test endpoint selection before a single patient is enrolled. Natural language processing pipelines process real-world evidence alongside trial data in parallel, dramatically compressing time-to-insight. Adaptive trial designs — where randomization ratios or dosing arms adjust based on incoming efficacy signals — are increasingly enabled by AI inference engines running against streaming clinical data. The patient experience is now a measurable competitive differentiator, with AI reducing participant burden through remote data collection, predictive engagement monitoring, and automated scheduling.

Research & Science

AI Finds Drug Safety Signals in Clinical Notes — Cutting Manual Review Time Across Multi-Site Cancer Immunotherapy Trials

Executive BriefA new multicenter study published in eBioMedicine shows AI can detect dangerous immune-related drug reactions — the kind that can be fatal in cancer immunotherapy patients — directly from clinical notes, potentially replacing costly and time-consuming manual chart reviews across trial sites.

The study tested zero-shot learning with GPT-3.5, GPT-4, and GPT-4o to identify immune-related adverse events (irAEs) in unstructured clinical notes. The prompt specified six immune checkpoint inhibitors (ICIs) and dozens of associated irAEs, with no training examples. GPT-4o achieved the best performance: patient-level F1 scores of 56% at Vanderbilt Health (100 patients), 66% at UCSF (70 patients), and 62% across Roche trial notes (272 patients). Lead author Cosmin Bejan, PhD, noted that even this moderate accuracy level could significantly reduce the per-patient resource burden of pharmacovigilance when applied at scale across multi-site ICI trials.

First Large-Scale RCT of AI Early-Warning System Finds Passive Display Alone Doesn't Reduce Clinical Deterioration

Executive BriefThe largest randomized controlled trial of an AI deterioration-prediction tool to date found a sobering result: showing AI predictions to clinical staff on a passive dashboard did not significantly reduce patient deterioration events. The finding doesn't invalidate AI early-warning systems — it reveals that implementation design, not just algorithm performance, determines real-world impact.

The pragmatic RCT enrolled 10,422 inpatient visits across medical and surgical units. Patients were randomly assigned to care teams with or without access to a passive AI-based predictive analytics display. The primary endpoint was hours free of clinical deterioration events. The trial represents one of the first rigorous applications of the RCT gold standard to an ambient AI monitoring tool in a real-world inpatient environment. Results suggest that meaningful clinical integration — not mere data availability — is required for AI predictions to translate into improved outcomes.

Policy & Regulation

Health Systems Should Prepare Now for Increasing AI Enforcement — Before the First Subpoena Arrives

Executive BriefRegulatory enforcement of healthcare AI has left the theoretical stage. Federal agencies and state attorneys general are already applying existing statutes to AI conduct — and health systems that haven't audited their AI tools for HIPAA, FTC, and consumer protection compliance face real legal exposure right now.

The FTC is applying Section 5 of the FTC Act to misleading AI capability claims, undisclosed AI use in clinical workflows, and data practices tied to automated decision-making. OCR has clarified that AI systems processing protected health information (PHI) — including ambient documentation tools, AI scribes, and prior authorization software — must meet the same HIPAA safeguard standards as any EHR or practice management system. State AGs are adding antitrust and false claims theories. The enforcement article recommends health systems immediately audit AI vendors for PHI data flows, review marketing representations, and document clinical oversight protocols.

Five States Pass Prior Authorization Reform Laws Targeting AI-Driven Coverage Denials

Executive BriefVirginia, Washington, North Dakota, Indiana, and Alaska enacted prior authorization reform laws in 2026, directly targeting three failure modes where AI is contributing to delayed or denied care. This is state-level regulation moving faster than any federal action.

The five state laws collectively address: (1) AI processing bottlenecks causing authorization delays beyond statutory timeframes, (2) repetitive prior authorization requirements for recurring treatments without documented clinical rationale, and (3) AI-generated denials issued without adequate human clinical review or appeal pathway. All five laws preserve human clinician authority over final coverage determinations when AI tools are used in the review workflow. The legislation arrives simultaneously with CMS's WISeR pilot — creating a complex federal/state regulatory patchwork for insurers operating in multiple markets.

AI Enforcement Accelerates as Federal Policy Stalls and States Step In

Executive BriefThe federal government is trying to preempt state AI laws through an executive order and a new AI Litigation Task Force — but states have already passed 240+ AI bills in 2026 alone and show no sign of slowing down. Healthcare AI companies now face a patchwork of 43 overlapping state compliance regimes.

43 states introduced over 240 AI-related bills in 2026, matching the entire prior year's legislative output in just the first quarter. Emerging state themes include: mandatory patient disclosure of AI use in clinical care, clinical oversight requirements for AI diagnostic tools, prohibitions on AI-only coverage denial decisions, and AI sandbox programs enabling supervised deployment. California's AB 489 (effective January 1, 2026) prohibits AI from impersonating licensed healthcare professionals, with enforcement routed through licensing boards. The White House's AI Litigation Task Force, directed to challenge state laws inconsistent with federal policy, is expected to trigger constitutional preemption litigation in multiple jurisdictions.

Beyond Detection: In the Age of Clinical AI, What Counts as an FDA 'Breakthrough' Medical Device?

Executive BriefThe FDA's Breakthrough Device Designation was built for life-saving treatment innovations, but AI companies are now seeking it for predictive risk tools and workflow optimizers — forcing the agency to rethink whether the program's statutory requirements fit a world where the most impactful AI is preventive, not curative.

The FDA cleared 295 AI/ML-enabled devices in 2025 alone, with three in four clearances in imaging. The Breakthrough Device Designation provides expedited review and early FDA interaction for technologies offering more effective treatment or diagnosis of life-threatening conditions. AI tools for early detection, deterioration prediction, and operational optimization now seek this designation — but the statutory language, focused on treatment efficacy, creates ambiguity for prevention and workflow tools. The FDA is also updating Quality Management System Regulation (QMSR) in 2026 to align with ISO 13485:2016 international standards, affecting all AI device manufacturers.

AI Prior Authorization Pilot Hits Original Medicare — Covering Nearly 1 in 5 Beneficiaries in Six States

Executive BriefCMS launched a 5-year AI prior authorization pilot affecting nearly 20% of traditional Medicare beneficiaries — the first time algorithmic review has been embedded into original Medicare at scale. Patient advocates are alarmed; the data on AI-driven denials is not reassuring.

The CMS WISeR (Wasteful and Inappropriate Services Reduction) Model, launched January 2026 and running through 2031, applies AI-assisted prior authorization review to 17 service categories in Arizona, New Jersey, Ohio, Oklahoma, Texas, and Washington. Targeted services include knee arthroscopy for osteoarthritis, certain nerve stimulation procedures, and skin/tissue substitutes — categories CMS identified as vulnerable to fraud, waste, and inappropriate use. Final coverage decisions require licensed clinician sign-off, not AI alone. However, a 2024 Senate committee report found AI prior authorization tools are associated with denial rates 16 times higher than decisions made without AI — a finding that makes the CMS pilot a flashpoint for healthcare AI governance nationally.

Industry & Business

Digital Health Funding Hit $4B in Q1 2026 as AI Becomes Table Stakes — Best Quarter Since Late 2021

Executive BriefDigital health just had its strongest fundraising quarter in four years — $4 billion across 110 deals — powered by AI mega-rounds that are concentrating capital among proven platforms and leaving early-stage companies starved for oxygen. If you're not AI-native, you're not getting funded.

Rock Health data shows Q1 2026 digital health funding totaled $4 billion across 110 deals, up $1 billion from Q1 2025's $3 billion across 122 deals. Average deal size hit $36.7 million — the highest since Q4 2021. Twelve mega-deals ($100M+) accounted for 59% of all capital deployed. WHOOP led with a $575 million Series G; other major rounds included Verily ($300M), OpenEvidence ($250M), Talkiatry ($210M), and eMed. $2.88 billion specifically targeted disease-agnostic AI platforms, signaling that horizontal AI infrastructure — not condition-specific apps — is the dominant investment thesis.

AI Doctor Startup Doctronic Raises $40 Million to Scale a Model Where AI Does the First 80% of a Clinical Encounter

Executive BriefDoctronic raised $40 million to build out an AI-first primary care model where the AI handles intake, history-taking, and preliminary clinical assessment — then hands a structured summary to a physician for review and final decision. It's a bet that AI can extend physician reach rather than replace it.

Doctronic's platform uses large language models trained on clinical guidelines, diagnostic reasoning frameworks, and structured medical interviewing protocols to conduct patient intake, generate differential diagnoses, and surface relevant evidence. The AI completes the first ~80% of the clinical encounter workflow, producing a structured assessment that a reviewing physician can validate and act on. The $40 million in funding will expand the platform's specialty coverage and enterprise health system integrations. The model addresses physician capacity constraints rather than displacing clinical judgment — an increasingly common framing for AI-first care delivery startups seeking provider partnerships.

OpenAI Launches ChatGPT Health — Connecting Medical Records and Wellness Apps in a Personal AI Health Co-Pilot

Executive BriefOpenAI launched ChatGPT Health — a dedicated, encrypted health experience inside ChatGPT that connects to users' medical records, Apple Health, and wellness apps. For the first time, a general-purpose AI can reason over an individual's complete health data longitudinally, shifting the consumer AI health conversation from generic information to personalized context.

ChatGPT Health integrates with Apple Health, MyFitnessPal, and Function through purpose-built encrypted data pipelines. Health conversations are compartmentalized from the rest of ChatGPT and explicitly excluded from model training data. Users can connect patient portals to interrogate lab results, prepare for clinical appointments, and model insurance plan tradeoffs based on their own health patterns. OpenAI deployed layered encryption and memory isolation specifically for health data, and explicitly disclaimed the feature for clinical diagnosis or treatment decisions. Initial rollout targeted Free, Go, Plus, and Pro users outside the EEA, Switzerland, and UK, with full web and iOS availability planned in subsequent weeks.

The American Powerplay: U.S. Captures 76% of Global Digital Health Investment in Q1 2026

Executive BriefThe United States isn't just leading global digital health investment — it's running away with it. Three in four investment dollars in digital health went to U.S. companies in Q1 2026, widening a gap that reflects the U.S.'s concentration of AI infrastructure, enterprise health system clients, and frontier model capabilities.

Galen Growth's Q1 2026 analysis shows the U.S. captured 76% of global digital health funding — a concentration reflecting the flywheel dynamics of AI infrastructure investment. Companies with proven enterprise health system contracts, demonstrated clinical ROI, and AI platform architectures attracted disproportionate capital versus early-stage or non-AI-native competitors. The analysis aligns with Rock Health data showing early-stage deal activity declining even as total capital rose, suggesting investors are making larger bets on fewer, more established companies rather than funding broad experimentation.

Social Buzz

Americans May Be Losing Trust in Healthcare AI — Even as Half the Country Already Uses It for Medical Decisions

Executive BriefA national survey found trust in healthcare AI dropped 10 percentage points in two years — with only 42% of Americans now open to AI in their care. The uncomfortable twist: 51% of adults have already used AI to make an important health decision without consulting a doctor. Trust is falling while use is rising.

The survey found 42% of U.S. adults open to AI participation in their healthcare, down from 52% in the 2024 baseline measurement. Simultaneously, 51% report having used AI to make a significant health decision without consulting a medical professional — a 9-point gap between stated trust and actual behavior. The divergence mirrors patterns seen in other technology adoption curves and signals a credibility crisis specific to healthcare AI: high utility perception combined with low institutional trust. For health systems, the data suggests patient AI tools need explicit accuracy communication and clear disclosure of limitations to close the trust gap.

As AI Makes More Health Coverage Decisions, the Risks to Patients Grow

Executive BriefThe same week that executives from nearly every major health insurer declared AI coverage review tools are saving money, the Trump administration expanded AI's role in Medicare prior authorization — and patient advocacy groups are warning that the combination is systematically eroding Americans' access to care.

The article documents converging pressures: commercial insurers publicly framing AI review as an efficiency tool, CMS's WISeR pilot embedding AI in original Medicare prior auth, and documented evidence that AI tools produce denial rates 16 times higher than human-only review. Louisiana and Maine are cited as enacting laws with explicit AI disclosure requirements in clinical encounters. The tension is creating a policy flashpoint between a federal administration openly supportive of AI cost-reduction applications and state-level legislators focused on patient access harm. The article circulated widely among healthcare policy practitioners and patient advocate networks in the week of publication.

Can AI Fix the Mental Health Crisis? Clinicians Are Cautiously Optimistic — With a Long List of Caveats

Executive BriefThe honest answer is: not yet — and probably not in the way people imagine. AI is delivering real, measurable gains in the administrative infrastructure of mental health care. It is not yet delivering validated clinical therapy. The two are very different things, and conflating them is the current risk.

AI tools proven to work in behavioral health in 2026 are predominantly operational: routing referrals, predicting no-shows with 70%+ accuracy, flagging medication non-compliance, and ambient documentation tools that reduced physician burnout rates from 51.9% to 38.8% in a JAMA Network Open multicenter RCT after just 30 days. Clinical AI for therapy delivery — tools that conduct or assist actual therapeutic interactions — remain in early-stage validation without sufficient RCT evidence to support broad deployment. Prominent psychiatrist Dr. John Torous of BIDMC noted that "we're not seeing a lot of clinical use of AI today," and cautioned that well-resourced tools remain expensive to operate at scale. The emerging consensus model: AI for homework, skills practice, and real-time session feedback; licensed clinicians for the therapeutic relationship.