// healthcare
AI in Healthcare & Life Sciences
AI runs the healthcare back office and the exam-room microphone, not the diagnosis — and new liability rules now follow the money.
standing summary updated November 6, 2026
By The Bleeding Edge AI desk. Drafted by AI from the week's linked sources and published automatically, without line-by-line human review. How we make this →
Where AI in healthcare and life sciences actually stands
Healthcare AI is deployed commercially in administrative and interface software, not in diagnosis or dosing. The largest surface is the back office: billing, coding, prior authorisation and the paperwork between a clinician and getting paid. Commure, a $7 billion startup, is built around automating that work, and R1, a revenue cycle processor handling over 600 million transactions a year, has agreed to buy Humata Health for AI-powered prior authorisations. The second surface is the patient portal: Oracle Health's AI-backed portal is generally available in the US, runs on OpenAI foundation models and is wired into Oracle's EHR. The third is documentation inside the encounter. Epic's Ergo Visit compiles outpatient summaries, with Ochsner Health the first system to implement it, and Teladoc ships an AI scribe as a built-in feature of its virtual care platform; Suki distributes through platforms including Optum, Zoom and athenahealth and has raised a $90M Series E to build agents that handle orders and schedule follow-ups, moving documentation from recording care toward acting on it. A Doximity-sponsored test this week gives one reason the diagnosis-and-dosing line still holds: it found about 19.5 of 1,000 medication-dosing answers from nine general-purpose AI models carried a serious or potentially lethal error. The study shipped alongside Doximity's own dose-checking product, so the figure is a claim to test rather than a settled benchmark, but no system in our coverage lets a general-purpose model set a dose directly, and this week's data is consistent with why. Inference
The back office is also where healthcare AI's economics are contested, and the argument runs against the usual pitch. CMS Administrator Mehmet Oz has said AI will be inflationary in the short term because it makes existing billing systems work more effectively, and researchers have observed rising costs tied to AI in coding and billing. The Blue Cross Blue Shield Association says payers spent about $942 million more over two years because of AI-backed coding, documentation and ambient tools, without a matching rise in treatment. That figure comes from payers alone, billing companies dispute it, and no one has independently checked it. Read together, these claims describe one mechanism: software sold to providers as a way to capture missed revenue shows up for payers as higher spend. Health systems whose AI business case rests on cost reduction should test it against coding-driven spend, and those using autonomous coding should keep records of why codes changed. Inference
Regulation is pulling in two directions at once. At the federal level, ONC's HTI-5 final rule cuts 14 of the 60 criteria in its Health IT Certification Program and ends the requirement that developers provide training data about AI-enabled tools, with regulators citing $1.5B in reduced regulatory costs — certification now tells buyers less, pushing training-data questions into each health system's own procurement review. Inference Against that loosening, the AMA has now put a number on who answers when clinical AI causes harm: developers for safe design, health systems for tool selection, training and monitoring, and physicians for clinical judgment, giving procurement and compliance teams a defined share of legal exposure they didn't clearly have before. That liability question is colliding with a jurisdictional one — STAT reports FDA and CMS are working behind the scenes to head off a growing state-federal clash as more states test AI in clinical settings, which matters most for Utah, whose Office of AI Policy has approved a pilot letting a Nolla Health dermatology app prescribe and refill topical acne drugs, with physicians reviewing every prescription at first and later only after the fact; doctors have pushed back and Utah has reported no outcomes. If federal rules end up overriding state-approved pilots, health systems that joined a sandbox could find the ground shifting under deployments they already built. Inference Patients, meanwhile, want more oversight than the deployment trend is giving them: a Pew Research Center survey of 3,488 US adults found people are open to AI handling scheduling and ambient scribing but still want a human in the loop even there, and are far less comfortable with AI diagnosing conditions or adjudicating claims — comfort tracks the risk of the task, not what the model can technically do. Inference A separate Wolters Kluwer/Ipsos survey found about 89% want human oversight of AI health answers, and Pew has also found nearly three-quarters of US adults say providers should disclose AI use, with nearly half unsure whether AI has already touched their care; as federal disclosure requirements shrink, that burden falls on providers rather than on certification. Inference
At the research end, AI drug design has turned into dealmaking, but deal structure shows how little has been proven. Zealand Pharma has signed a deal worth up to $1.3B to develop protein-based obesity and metabolic treatments with Generate Biomedicines' AI, with only $30M, about 2%, upfront and the rest tied to milestones. Chai Discovery has a deal with GSK to make biologics with AI, following its Eli Lilly collaboration, with no terms reported. Nvidia-backed Iambic Therapeutics has an undisclosed R&D partnership with AbbVie and has filed to go public, though a filing is not a listing. Pharma is paying for options on AI-designed molecules rather than for results, and the first real test will be whether any candidate reaches the clinic fast enough to earn its milestones. On the current record, AI in healthcare mostly administers, bills and explains care; its one sanctioned step into delivering care is a narrow state pilot with no outcomes yet, a new AMA framework now spreads liability for that step across developers, deployers and physicians, and the rules governing all of it are getting lighter at the federal level even as patients ask for more disclosure, not less. Inference
// Big moves
2026-W45The AMA says liability for clinical AI harm splits three ways: developers answer for safe design, health systems (deployers) for tool selection, training and monitoring, and physicians for clinical judgment — the first clear framework spreading legal exposure beyond doctors alone.
CorroboratedHealthcare Divesource →
STAT reports FDA and CMS officials are working to head off a brewing state-federal conflict as more states stand up their own AI sandboxes for clinical testing, raising the risk that federal rules could later override state-approved pilots.
UnverifiedSTATsource →
A Pew Research Center survey of 3,488 US adults found patients are open to AI handling scheduling and ambient scribing but still want a human involved, and are far less comfortable with AI diagnosing conditions or adjudicating claims.
CorroboratedHealthcare Divesource →
Health tech company Doximity says nine widely used AI models gave a seriously or potentially lethally wrong medication-dosing answer about 19.5 times per 1,000 questions, releasing the figure alongside its own dose-verification product.
UnverifiedHealthcare Divesource →
// Daily stories
2026-11-05
Pew: patients want a human in the loop, even for AI scheduling and scribingCorroboratedHealthcare Dive2026-11-04A Pew Research Center survey of 3,488 US adults found patients are open to AI handling scheduling and ambient scribing, but want a human involved, and are far less comfortable with AI diagnosing conditions or adjudicating claims.
why it matters
Patient comfort tracks task risk, not AI capability — health systems expanding AI should prioritize administrative workflows first and keep humans explicitly visible in anything clinical.
2026-11-03
AMA: physicians, AI vendors and health systems share liability for clinical AICorroboratedHealthcare Dive2026-11-02The AMA says physicians, AI developers and healthcare deployers each bear distinct responsibility for clinical AI harm: developers for safe design, deployers for tool selection, training and monitoring, physicians for clinical judgment.
why it matters
Liability for clinical AI failures no longer rests on physicians alone — health system leaders now need procurement and monitoring that can withstand shared-liability scrutiny.
2026-10-31
FDA and CMS move to head off a state-federal clash as more states test AI in clinical settingsUnverifiedSTAT2026-10-30STAT reports that FDA and CMS officials are working behind the scenes to head off a growing conflict between the states and the federal government over regulating AI. More states are looking to test AI in clinical settings.
why it matters
Health systems thinking about joining a state AI sandbox should watch whether federal rules end up overriding state approvals. That risk matters before committing clinical staff and budget to a pilot.
2026-10-30
Doximity study: about 19.5 in 1,000 AI dosing answers had a serious or potentially lethal errorUnverifiedHealthcare Dive2026-10-29Health tech company Doximity says that when nine widely used AI models were asked about medication dosing, about 19.5 in 1,000 answers had a serious or potentially lethal error. Doximity released the research alongside its own dose verification tool.
why it matters
This study comes from a vendor and was released with its own product, so treat the figures as a claim to test. If you're piloting general-purpose models in clinical workflows, dose checking is a concrete place to ask for evidence.
2026-10-27
ONC's HTI-5 final rule cuts 14 of 60 health IT certification criteria and ends AI training-data disclosureCorroboratedHealthcare Dive2026-10-27ONC finalized HTI-5, which removes 14 of the 60 criteria in its Health IT Certification Program and revises seven more. Developers will no longer have to provide training data about tools that use AI. Regulators say the changes cut $1.5B in regulatory costs.
why it matters
Certification will now tell buyers less about the AI inside certified products. Health systems that relied on those disclosures may need to ask vendors about training data directly during procurement and governance reviews.
2026-10-16
AMA publishes a governance framework for agentic AI, setting four questions on tasks and accountabilityCorroboratedHealthcare Dive2026-10-15The American Medical Association released a governance framework for using AI agents in healthcare. It says agents performing clinical functions must be backed by evidence and maintain high-quality performance, and poses four questions on which tasks they may do, when, and who is accountable.
why it matters
Health systems already running agents for documentation, diagnostics or patient monitoring can use the four questions as an audit checklist now. Worth watching whether payers or regulators adopt the framework.
2026-10-15
Zealand Pharma signs up-to-$1.3B deal to design obesity drugs with Generate Biomedicines' AIUnverifiedEndpoints News2026-10-14Zealand Pharma will develop new protein-based treatments for obesity and other metabolic diseases using Generate Biomedicines' AI technology, in a deal worth up to $1.3B. The Danish drugmaker is paying $30 million upfront.
why it matters
$30M upfront against a $1.3B headline is about 2%, so almost all of the value in this AI drug-design deal depends on milestones. Worth watching whether AI-designed candidates reach the clinic fast enough to earn that structure.
2026-10-14
AI scribe maker Suki raises $90M Series E to build agents that file orders and schedule follow-upsUnverifiedHealthcare Dive2026-10-13CEO Punit Soni said Suki will use the funding to develop AI agents for tasks such as filling orders and scheduling follow-ups. It will also scale partnerships with health tech firms that embed its tools, including Optum, Zoom and athenahealth.
why it matters
Scribes are moving from taking notes to taking actions in clinical workflows, a shift Endpoints is also tracking. For health systems, the open question is who reviews an order an agent places before it goes through.
2026-10-10
Chai Discovery has signed a deal with GSK to make biologics with AI, Endpoints News reports. It follows the startup's collaboration with Eli Lilly, which Joshua Meier pitched at the start of 2026 as just the beginning.
why it matters
Another big-pharma partnership after Lilly suggests AI biologics design is moving into dealmaking. Worth watching whether terms or milestones surface that show how GSK values the work.
2026-10-08
Teladoc adds an AI scribe and contactless vitals monitoring to its virtual care platform for hospitalsUnverifiedHealthcare Dive2026-10-07Teladoc Health is adding two AI features to its virtual care platform for hospitals. Its AI scribe, which records and transcribes clinical encounters to create notes, is available now. Contactless vitals monitoring is in private preview with select customers, with broad release expected in December.
why it matters
Ambient scribing now comes built into an established virtual care platform, not only as a standalone product. Health systems on Teladoc should compare it with their existing scribe contracts, and look for accuracy data from the vitals preview before December.
2026-10-07
Utah approves more AI prescribing pilots, including an acne app that moves to after-the-fact reviewCorroboratedHealthcare Dive2026-10-06Utah's Office of AI Policy approved a new slate of healthcare AI pilots on Friday. One, with Nolla Health, tests whether a dermatology app can prescribe and refill a short list of topical acne drugs. Physicians first review every prescription, later only after the fact.
why it matters
The key step is moving from checking prescriptions before they go out to checking them afterwards. Providers and payers should watch what outcomes Utah reports before other regulators decide whether to follow, given pushback from doctors.
2026-10-06
Utah plans to let an AI product evaluate patients and write new prescriptions without prior human reviewUnverifiedSTAT2026-10-05Utah is adding more health AI pilots, covering prescriptions and women's health, STAT reports. The state plans to let an AI product evaluate patients and write new drug prescriptions without prior human review.
why it matters
Without prior review, oversight moves from each prescription to the product itself. Health systems and pharmacies should watch how Utah audits outcomes and handles accountability before other states copy the model.
2026-10-02
Wolters Kluwer survey: privacy fears hold patients back from healthcare AI; 89% want human oversightUnverifiedHealthcare Dive2026-10-01A Wolters Kluwer survey run by Ipsos finds patients are less willing to use AI for healthcare when it needs personal data or PHI. About 89% say AI health answers need human oversight. Under a third of US adults use AI to find a provider, navigate insurance or plan a diet.
why it matters
For providers and payers rolling out patient-facing AI, consent, data handling and visible clinician review look like requirements for adoption, not optional compliance work.
2026-10-01
STAT reports that HHS announced new efforts to speed up and expand clinical trials with AI. Federal health officials say they want faster clinical research in order to compete with countries such as China.
why it matters
Federal backing for AI in trials could affect site selection, recruitment and trial timelines for sponsors. Watch for program details: funding size, eligibility, and whether FDA guidance follows.
2026-09-29
BCBSA says AI billing tools cost it about $942M over two years; billing companies disagreeUnverifiedHealthcare Dive2026-09-28Research from the Blue Cross Blue Shield Association puts the cost to it of AI-backed coding, documentation and ambient tools at about $942 million over two years. BCBSA says spending rose without a matching rise in treatment. Billing companies dispute the finding.
why it matters
Health systems using autonomous coding should expect payers to look harder at how complex their coding has become. Worth watching whether this turns into audits or contract terms. Records that show why codes changed are worth keeping.
2026-09-25
CMS administrator Oz says AI will inflate healthcare costs before it lowers themCorroboratedHealthcare Dive2026-09-24At Oracle's annual health and life sciences summit, CMS Administrator Mehmet Oz said AI will be inflationary in the short term because it will turbocharge the ability of current billing systems to work more effectively. Researchers have observed rising costs tied to AI in coding and billing.
why it matters
The official who runs Medicare payment is signalling that AI's first effect is higher billing throughput. ROI cases built on cost reduction should be stress-tested against coding-driven spend and likely payer scrutiny.
2026-09-23
Nvidia-backed AI drug discovery firm Iambic Therapeutics files to go publicCorroboratedEndpoints News2026-09-22Iambic Therapeutics has filed to go public, per Endpoints News, which frames it as the latest example of maturation among AI-focused biotech startups. The San Diego company disclosed new details on its own pipeline programmes.
why it matters
An IPO filing forces the disclosure AI drug discovery has mostly avoided: pipeline stage, spend, and what the models have actually produced. For teams evaluating AI partners, it is a rare set of comparable numbers.
2026-09-22
AbbVie signs an R&D partnership with Iambic as the AI drug-discovery firm ships a new Enchant modelUnverifiedEndpoints News2026-09-21Iambic Therapeutics signed an R&D partnership with AbbVie and released the next generation of Enchant, one of its main AI models, Endpoints News reported. The San Diego company was founded in 2019. No terms were given.
why it matters
Large pharma keeps reaching for AI discovery capability through partnership rather than building it in house. With no terms disclosed, the signal is the pattern, not the price.
2026-09-11
ARPA-H kicks off ADVOCATE, putting millions into health tech firms building AI agentsUnverifiedEndpoints News2026-09-10ARPA-H, the federal agency that funds breakthrough health technologies, has kicked off a programme called ADVOCATE, which provides millions of dollars to health tech companies to build AI agents, Endpoints News reported.
why it matters
Federal money is now underwriting agentic tooling in health, which moves some build risk off provider and sponsor balance sheets. Worth checking whether ADVOCATE's scope overlaps work you were planning to fund internally.
2026-09-09
Anthropic posts for a biopharma corporate development lead to run acquisitions and partnershipsCorroboratedEndpoints News2026-09-08Anthropic is hiring a corporate development lead in biopharma to "source and execute acquisitions, investments, and strategic partnerships," according to a job posting reported by Endpoints News, which reads the listing as confirming the company's biotech dealmaking ambitions.
why it matters
A frontier lab buying into life sciences, rather than only selling models to it, would change the partner and acquirer landscape. It is one job posting — but worth watching what it actually buys.
2026-08-28
Pew: nearly three-quarters of US adults want providers to disclose when AI is used in their careCorroboratedHealthcare Dive2026-08-27A Pew Research Center survey published this week found nearly three-quarters of US adults say it is extremely or very important that providers disclose AI use, while 53% say they have little or no say over it and nearly half are unsure whether it has already been used in their care.
why it matters
Patient disclosure is drifting from a compliance checkbox toward a consent expectation. Worth deciding now how ambient scribes and imaging triage get surfaced to patients — the survey suggests most do not know it is happening.
2026-08-27
Sword Health to buy Headspace all-cash, per a state filing, with the close set for Sept 14CorroboratedHealthcare Dive2026-08-26Headspace parent OrangeDot told Massachusetts regulators on 22 July that it is seeking to be acquired by Sword Health in an all-cash deal expected to close on 14 September, Healthcare Dive reports. Terms were not disclosed and neither company commented.
why it matters
Employers and payers buying point solutions may find their virtual care and mental health vendors merging into one contract. If you hold both, worth reviewing renewal and data terms before the close date.
2026-08-25
FDA's digital health policy head says generative AI device guidance is comingUnverifiedSTAT2026-08-24STAT reports the FDA expects to issue guidance on its policies for generative AI devices, according to Rick Abramson, the agency's head of digital health policy. No timing or scope was given.
why it matters
Teams building generative AI tools still have no FDA guidance to design against. Worth watching what scope the agency defines before committing to a submission pathway or a launch date.
2026-08-21
R1 to buy Humata Health to automate insurer prior authorisations; terms undisclosedCorroboratedHealthcare Dive2026-08-20Revenue cycle firm R1 has agreed to acquire Humata Health, which specialises in AI-powered prior authorisations, with the deal expected to close in Q3. Humata's team will then integrate its technology into R1's revenue cycle platform. An R1 spokesperson declined to share financial terms.
why it matters
Prior authorisation is where provider revenue stalls. A processor handling over 600 million transactions a year buying the capability rather than building it is a make-or-buy marker for anyone budgeting revenue-cycle automation.
2026-08-20
Epic launches Ergo Visit for outpatient clinicians, with Ochsner Health the first to implementCorroboratedHealthcare Dive2026-08-19At its annual meeting in Wisconsin, Epic launched Ergo Visit, a tool that compiles patient information from its EHR and other sources into a summary for outpatient clinicians. Ochsner Health is the first health system to implement it, and five physicians outside the system are piloting it.
why it matters
Anything Epic ships lands on every health system's roadmap by default. The question for CIOs is whether outpatient summarisation displaces a point solution you already bought, and what evidence Epic offers on accuracy.
2026-08-13
Oracle Health's AI patient portal, built on OpenAI models, goes generally available in the USCorroboratedHealthcare Dive2026-08-12Oracle said Wednesday its AI-backed patient portal is now generally available in the US. Built on OpenAI foundation models and integrated with Oracle's EHR, it lets patients review records, engage providers, schedule appointments and get AI-generated explanations of medical jargon.
why it matters
Patient-facing generative AI is arriving as an EHR feature rather than a pilot, which shifts the question from whether to deploy to who reviews what patients are told. Contract terms decide whether you already own it.
2026-08-12
STAT reports from inside Commure, a $7 billion startup automating healthcare's administrative tasks, on what it calls the messy realities of the effort. Article is behind STAT+.
why it matters
Administrative automation, not diagnosis, is where healthcare AI revenue is concentrating. It is also where the failure modes are least visible to patients.
// Track AI in Healthcare & Life Sciences
One email a week on what actually changed — in healthcare & life sciences and across the frontier. Verified, labelled, no hype.