From TekinHealthHealthcare Tech Field Notes

The Cigna Group announced a partnership with OpenAI to build AI into clinical workflows at Cigna Healthcare and its Accredo Specialty Pharmacy, starting with cancer care. The first tool gives oncology nurses and case managers an AI-enabled interface that pulls together a patient's clinical, pharmacy, behavioral and benefits data into one view, along with patient-reported information like side effects and preferences. Cigna says the goal is to help clinical teams spot needs and coordinate support between physician visits, under human and clinician oversight. The companies plan to expand the collaboration to other conditions over time.

Why it's notable: It puts a large national payer's AI push directly into oncology case management workflows that clinics and specialty pharmacies already touch, rather than just member-facing chat.

OpenEvidence released a family of medical AI search models on September 3. Three are free to verified clinicians on its website and iOS and Android apps: Osler answers in about five seconds for questions during a visit, Sackett runs a deeper search in about 30 seconds, and Snow does a full literature review in about five minutes. A fourth model, Darwin, is available by application only to research partners such as the National Organization for Rare Disorders. The company says Darwin is the first model to score 100% on the MedQA benchmark and that it will add Darwin's capabilities to the three clinician models as safeguards are validated.

Why it's notable: OpenEvidence is one of the AI reference tools doctors use most. It now lets clinicians choose between fast answers at the point of care and slower, deeper evidence reviews. The benchmark results are the company's own claims and measure exam-style questions, not patient outcomes.

OpenAI announced on September 1, 2026 that ChatGPT for Healthcare now connects to Epic electronic health records, letting clinicians at participating organizations pull a patient's chart into the chatbot for read-only reference. Access requires an organization-level workspace covered by a signed business associate agreement rather than an individual account, and OpenAI says it evaluated the feature with 4,363 physician ratings across 27 clinical use cases, reporting a 99.1% "safe" rate. UCSF Health is OpenAI's named pilot partner for the Epic integration. ChatGPT for Healthcare itself launched in January 2026 with organizations including HCA Healthcare, Cedars-Sinai, Memorial Sloan Kettering, Boston Children's Hospital, AdventHealth and Stanford Medicine Children's Health.

Why it's notable: It is one of the largest sanctioned deployments of a general-purpose chatbot directly against live EHR data, and the roughly 1-in-100 "unsafe" response rate OpenAI itself reports means clinicians should treat its output as a starting point, not a verified answer, on charts that include allergies and medication changes.

The FDA's Digital Health Center of Excellence published a discussion paper on how it might regulate medical devices that use generative AI. The paper sets out a two-axis risk framework: how independently a device acts, and how much harm a wrong output could cause. It also proposes competency-based premarket testing that combines benchmarking with clinical confirmation, and it covers postmarket monitoring, foundation models and agentic systems. Comments are due October 19, 2026, under docket FDA-2026-N-7874. The FDA has not yet authorized any generative AI device.

Why it's notable: This is the clearest signal so far of how the FDA may treat LLM-based clinical tools, including chatbots and agents. Those rules will decide which of these tools can be marketed for clinical use and what evidence vendors must show.

At its 2026 Users Group Meeting in Verona, Wisconsin, Epic introduced Agent Factory, a no-code platform for health systems to build, run and monitor AI agents that act across Epic workflows. Organizations can start from about 120 built-in AI features or build their own. Epic says early adopters are using it now, with broader availability in 2027. Epic also announced Curiosity, a predictive tool built on its Cosmos clinical database; real-time prior authorization checks; and UpToDate content inside its clinician assistant, Art.

Why it's notable: Epic is the record system for most large US health systems, so its agent platform will shape how AI automation reaches everyday clinical and administrative work. For many organizations, it will also affect whether separate AI vendors are needed at all.

Twelve US health systems announced a Diagnostic AI Consortium with Aidoc on August 11, 2026. Members are Advocate Health, Cedars-Sinai, Hartford HealthCare, Houston Methodist, Mercy, Mount Sinai, Northwell Health, Northwestern Medicine, Sutter Health, UF Health, University Hospitals of Cleveland and WellSpan Health, which together care for nearly 20 million patients a year. The group plans to co-design AI-assisted diagnostic workflows, measure their effect on safety, quality and time to diagnosis across member sites, and agree on shared evaluation and governance practices. Aidoc is supplying the technology through its CARE foundation model and aiOS platform, and the group expects to share its first results in 2027.

Why it's notable: Health systems usually evaluate imaging AI one site at a time. Here a dozen large systems are pooling evidence on whether it actually speeds diagnosis, which could set the standards other hospitals use when they buy and govern these tools.

NHS England announced on July 6, 2026 that it will add an AI triage tool to the NHS App. The tool asks patients adaptive questions and directs them to a GP appointment, a pharmacy, A&E, community services or self-care. It is due to reach more than 200,000 patients in its first 12 months and all NHS App users by April 2028, and patients can still contact their practice the usual ways. NHS England says a pilot at a Sussex GP practice cut phone queues by 29%. The announcement is part of a £10 billion, three-year technology program that also expands AI notetaking for staff.

Why it's notable: It is one of the largest national deployments of AI front-door triage in primary care, and a test of whether automated routing can ease the morning phone rush at GP practices at scale.

On June 25, 2026, UpDoc publicly announced the FDA 510(k) clearance (K253281, granted in December 2025) of UpDoc V1.0. It is prescription software that talks with adults with type 2 diabetes by voice or text and adjusts their insulin within limits set by their physician. The company calls it the first FDA-cleared device to use patient-facing large language models. It says the system is deployed at Cleveland Clinic, Allegheny Health Network and UCSF Health, can order follow-up labs and writes its actions into the EHR. It was cleared as a 510(k) with an insulin dosing program as the predicate device, and it is backed by a small Stanford randomized trial published in JAMA Network Open. STAT reported that the CEO would not say whether the language model itself makes the dosing decisions.

Why it's notable: This is the first regulatory path for an AI that talks directly with patients to adjust a drug between visits, and other developers are likely to follow it. It also leaves open how much of the clinical decision the language model makes and how much fixed dosing rules make.

Assort Health raised a $120 million Series C led by Menlo Ventures at a $1.2 billion valuation, announced June 24, 2026, bringing its total funding to more than $222 million. Its voice and text agents answer patient calls and handle scheduling, triage, referrals, intake forms, refills, eligibility checks and payments, and they write encounter details back to the EHR. The company says revenue grew 20-fold over 15 months and that John Muir Health saw a 5% absolute increase in net appointment volume after deploying it.

Why it's notable: AI agents answering the front-desk phone are becoming one of the fastest-growing uses of AI in clinics and specialty practices, and they directly affect access, no-shows and staffing.

On April 23, 2026, OpenAI launched ChatGPT for Clinicians, a version of ChatGPT for clinical work such as documentation, medical research and care questions, free to US physicians, nurse practitioners, physician assistants and pharmacists who verify their NPI and license. Responses cite journal sources, and eligible questions can earn CME credit. OpenAI also released HealthBench Professional, an open benchmark for clinician chat tasks, and says that before launch it tested nearly 7,000 real clinical conversations, with 99.6% of responses rated accurate and safe.

Why it's notable: A free, citation-backed AI assistant from a major AI lab competes directly with paid clinical reference and search tools. Many clinicians will now have it within easy reach.

On April 16, 2026, Hippocratic AI launched Nurse Co-Pilot, an AI voice assistant for inpatient nurses built with Cincinnati Children's Hospital Medical Center, OhioHealth and Cleveland Clinic, which are the first to use it. It starts with patient-facing tasks such as admission education, patient education, caregiver engagement and medication-related support, and the company says it can give nurses back 1 to 4 hours per shift. Hippocratic also launched an 'AI Front Door' voice agent that handles incoming patient calls.

Why it's notable: Patient-facing AI agents are moving beyond call centers onto hospital floors, taking over routine education and discharge conversations that nurses now spend hours on.

On April 15, 2026, ambient scribe company Abridge announced multi-year content deals with NEJM Group and the American Medical Association's JAMA Network. Content from those journals will feed Abridge's clinical decision support, where clinicians ask questions inside the EHR and get cited answers that draw on both published research and the patient conversation. Decision support using UpToDate content is already live, and Abridge said the NEJM and JAMA content would become generally available in the following months. Abridge says it serves about 250 large US health systems.

Why it's notable: Ambient scribes are adding evidence lookup, which puts them in competition with clinical reference tools.

Mount Sinai Health System announced on March 31, 2026 that it is rolling out OpenEvidence, an AI medical search tool, inside its Epic EHR across its seven hospitals. It is giving access to physicians, registered nurses and pharmacists. OpenEvidence answers clinical questions in plain language, with citations to peer-reviewed literature and guidelines. Coverage described it as OpenEvidence's first enterprise-wide deal with a health system.

Why it's notable: Clinicians have mostly used OpenEvidence on their own phones. This deal makes it a sanctioned tool inside the EHR for the whole care team, which is a model other health systems may follow to govern clinicians' use of AI.

On March 10, 2026, Amazon began offering Health AI, an agent-based health assistant built on Amazon Bedrock, on Amazon.com and in the Amazon app. It had launched for One Medical members in January. With the user's permission, it pulls medical history, medications, lab results and clinical notes through health information exchanges. It then answers questions, helps with prescriptions and scheduling, and hands users to One Medical clinicians. Amazon says eligible Prime members get up to five free message-based visits for more than 30 common conditions. After that, visits cost $29 each. Amazon also says auditor and escalation agents send uncertain cases to human clinicians.

Why it's notable: A large consumer platform now puts an AI agent that reads a patient's own records in front of well over 100 million Prime members, and connects it directly to cheap virtual visits. Primary care practices will increasingly see patients who have already talked to such an agent.

Before HIMSS26, Microsoft announced on March 5, 2026 that Dragon Copilot now connects to Microsoft 365 Copilot. A new Work IQ layer lets clinicians pull email, Teams, schedules and documents into the same assistant they use to write notes. Microsoft says the product can draft prior authorization requests from the chart and a payer's coverage criteria. It also opened a marketplace for third-party agents, with launch partners Optum, Regard, Humata Health and Canary Speech. The release also adds nursing flowsheet documentation, with a radiology integration in US preview. Microsoft says more than 100,000 clinicians use Dragon Copilot.

Why it's notable: One of the two biggest ambient scribe vendors is turning its note-taking tool into a general clinical assistant that takes on admin work like prior authorization. That changes what health systems are buying when they pick a scribe.

On February 4, 2026, Epic made AI Charting broadly available to its EHR customers. The tool listens during patient visits, drafts the clinician's note and queues up suggested orders from the conversation. Clinicians must review and sign off on all drafts and orders. Epic first previewed the tool at its August 2025 users group meeting. It is part of Epic's Art suite of clinician AI tools, and Epic says a chart-summary feature in that suite is now used more than 16 million times a month. Epic has not published a separate price for AI Charting.

Why it's notable: Epic runs the EHR at about 42% of US acute care hospitals, so a scribe built into the EHR puts direct pressure on standalone vendors like Abridge, Microsoft's Dragon Copilot and Ambience. It also gives Epic customers a ready alternative to buying a separate scribe.

On January 11, 2026, as the J.P. Morgan Healthcare Conference opened, Anthropic launched Claude for Healthcare, a set of HIPAA-ready AI tools for health systems, payers and patients. It includes built-in connections to the CMS Coverage Database, ICD-10 codes, the NPI Registry and PubMed, which Anthropic pitches for prior authorization, claims appeals, coding and care coordination. The company says the product runs on its Claude Opus 4.5 model. Record company HealthEx said it had become the first way for consumers to connect their health records to Claude, and Apple Health and Android Health Connect integrations were set to roll out in beta in the Claude mobile apps.

Why it's notable: A second major AI lab is now selling HIPAA-ready tools that connect directly to the Medicare coverage and coding data behind prior authorization and appeals, one of the biggest administrative costs for clinics and payers.

On January 7, 2026, OpenAI announced ChatGPT Health, a separate health space inside ChatGPT where US users can connect their medical records through the health data network b.well, which covers about 2.2 million US providers. Users can also connect apps including Apple Health, MyFitnessPal, Weight Watchers, Peloton and Function. OpenAI says more than 230 million people ask ChatGPT health questions each week and that the product was built with feedback from more than 260 physicians. The company says ChatGPT Health is not meant for diagnosis or treatment and that health conversations are not used to train its models. It launched first to a small group of early users, and the record connections are US-only. On January 12, OpenAI bought the health-records startup Torch to build out the product, for a price reported at $60 million to $100 million.

Why it's notable: The most widely used consumer chatbot can now read patients' own records, so more patients will arrive at visits with AI interpretations of their lab results and charts.

On January 6, 2026, FDA Commissioner Marty Makary announced at the Consumer Electronics Show that the agency was relaxing oversight of some digital health products. The same day the FDA issued revised final guidance on clinical decision support (CDS) software, replacing its 2022 version, and on low-risk general wellness products. Under the CDS guidance, the FDA will generally not regulate software that gives clinicians a single recommendation, such as a specific drug to consider, if the software meets its other non-device criteria. The wellness guidance lets non-invasive wearables report readings such as blood pressure and blood glucose without FDA review, as long as they are marketed only for wellness and make no medical claims. The guidances were issued without a prior public comment period.

Why it's notable: More AI decision-support tools and consumer wearables can now reach clinicians and patients without FDA review, so validating them falls more on the health systems and clinics that adopt them.

On December 1, 2025, the FDA said it had deployed agentic AI tools, which can plan and carry out multi-step tasks, to all of its employees. The agency listed uses including meeting management, premarket reviews, review validation, postmarket surveillance, inspections and compliance. Use is voluntary, and the models run in a secure GovCloud environment. The FDA says they do not train on staff inputs or on industry submissions. It also says all AI output is checked by staff before it goes into any regulatory action. The rollout builds on the Elsa generative AI tool launched in May 2025, which the agency says more than 70% of staff use.

Why it's notable: The regulator that clears drugs and devices is now putting AI agents into its own review work. That could affect how fast, and how consistently, applications from device makers and drug sponsors are handled.

Oracle Health said its new EHR, rebuilt from scratch on Oracle Cloud Infrastructure rather than on the Cerner platform, received ONC health IT certification and met DEA requirements for electronic prescribing of controlled substances. That makes it available to US ambulatory providers. The system includes Oracle's clinical AI agent and voice commands that let clinicians ask for things like recent labs and current medications instead of clicking through screens. Oracle said acute care functionality would follow in 2026.

Why it's notable: Outpatient practices that are choosing an EHR now have a certified option built with AI throughout, from one of the two largest US EHR vendors. It is also the first real test of whether Oracle can win back ground after the Cerner acquisition.

Hippocratic AI raised a $126 million Series C led by Avenir Growth at a $3.5 billion valuation. CapitalG, General Catalyst, Universal Health Services, Cincinnati Children's and WellSpan Health also took part. The company builds AI agents that talk with patients about scheduling, intake, chronic disease check-ins and post-discharge follow-up, and it does not let them diagnose or prescribe. It says it works with more than 50 health systems, payers and pharma companies in six countries, including Cleveland Clinic, Northwestern Medicine and Moffitt Cancer Center, and plans to use the money partly for acquisitions.

Why it's notable: Health systems are putting AI agents to work on routine patient phone and message outreach that staff used to handle. Investors are betting that this non-diagnostic work will scale first.

At HLTH 2025, Optum launched Optum Real, an AI-backed system that lets providers ask a payer in real time about a patient's benefits and whether a claim will be covered at the point of care, rather than weeks later during coding. The system uses AI to read payer contracts and policies. UnitedHealthcare is the first health plan on it, and Allina Health has been piloting it for outpatient radiology and cardiology claims, where it has processed more than 5,000 visits. Optum reported cuts in provider call volume of 25% to 42% and up to 75% fewer reimbursement submission errors in pilots.

Why it's notable: Claims denials and rework are a major cost for practices, and a payer-owned system that settles coverage questions up front could change how revenue cycle teams work, if payers beyond UnitedHealthcare adopt it.

OpenEvidence, an AI medical reference tool for clinicians, raised $200 million at a $6 billion valuation in a round led by Google Ventures, with Sequoia, Kleiner Perkins, Blackstone, Thrive, Coatue, Bond and Craft participating. The round came three months after a $210 million raise at a $3.5 billion valuation. The company says monthly clinical consultations nearly doubled to about 15 million since July. The tool is free for verified clinicians and paid for by advertising, and it draws on journals including JAMA and NEJM.

Why it's notable: OpenEvidence has become one of the most widely used generative AI tools among U.S. clinicians, and its ad-supported free model sets it apart from AI products that health systems license.

At HLTH 2025, Microsoft added an ambient AI capability for nurses to Dragon Copilot. It captures nurse-patient conversations at the bedside and turns them into flowsheet entries in the EHR, which the nurse reviews before they are filed. The feature runs inside Epic Rover, the mobile app nurses already use. Baptist Health and Mercy are early adopters, and Microsoft said general availability for U.S. nurses would begin in December 2025. Microsoft also opened Dragon Copilot to partner apps and agents from companies including Elsevier, Wolters Kluwer UpToDate, Cohere Health and Press Ganey.

Why it's notable: Most ambient AI so far has been aimed at physician notes. Nurses do much of the charting in a hospital, so a scribe built into their flowsheets and mobile workflow could relieve a large share of the documentation burden.