Issue 3
Viewpoint
regulation · United States
FDA discussion paper would assess generative AI devices in a way similar to doctors
On August 18, the FDA published a discussion paper on regulating generative AI-enabled medical devices: benchmarks would come first, then clinical confirmation, and finally evaluation of real-world use. Manufacturers, physicians and researchers can submit comments on document FDA-2026-N-7874 until October 19.
Who is affected and how
- medtech company
- During the comment period, companies should indicate which tier of the proposed graduated assessment their product would fall into and what evidence they can provide for it.
- doctor
- Through professional organizations, physicians can have a say in what clinical confirmation generative AI tools should be required to show.
- health IT
- Teams need to assess which embedded generative functions could fall under device regulation.
- health policy
- The FDA is signaling a direction for authorizing generative AI, which may serve as a benchmark for other regulators.
tool · United States
Epic UGM: agent-building tools, Cosmos-based generative models, real-time prior authorization
At its user meeting in Verona, Epic presented its new AI system, Ergo, and the Cosmos Curiosity family of generative models. Updates will now arrive quarterly (previously every 18 months), and the…
hospital managerBecause of Epic's AI cost model ('tokenomics'), CIOs need to plan a separate AI budget.health ITThe quarterly release cadence and agent-building tools require faster testing and governance processes.doctorAI features built into Epic may affect everyday documentation and decision-support work.health policyAccording to STAT, Epic is weighed down by internal tension over AI strategy and succession, and is also under scrutiny over market control.licensing · United StatesFDA has authorized more than 1,500 AI devices, but evidence is lagging
By mid-August the FDA's list included more than 1,500 AI-based medical devices, and the range of oncology applications is expanding. According to analyses, the evidence behind most devices has not…
doctorAuthorized status alone does not indicate clinical benefit, so the validation behind a purchased device should be asked about separately.hospital managerIn procurement, local or prospective evidence should be requested from the vendor; the FDA list alone is not enough.medtech companyGiven the evidence gap, companies should expect stronger demands for clinical substantiation.incident · United StatesAI scribe errors: invented patient data, missed social factors, language challenges
A report says 40 percent of doctors use AI scribes, but the tools can "hallucinate" patient data; in one case a patient's record listed psychedelic mushrooms she had never taken. A medRxiv study…
doctorGenerated notes must be reviewed before signing, especially medications and medical history.hospital managerAn error-reporting and sample-based audit process should be introduced for the quality of AI notes.health ITTool performance should be tested separately for multilingual patient groups.research · United States93% of hospitals use AI, but only 44% have tested it properly
A report finds that 93% of hospitals already run AI, but only 44% have tested it properly. The AMA and CHAI urge guardrails and responsible use.
hospital managerA formal process is needed to test and oversee AI tools.health ITA validation and monitoring framework must be built for the tools in use.doctorThe AMA urges guardrails for the use of AI in healthcare.Hungary
clinical deploymentPoland selects suppliers for all five areas of national AI diagnostics programmehealth policyThe Polish model is a regional benchmark for setting up a national AI diagnostics programme.doctorIn a nationwide system, the workflow for reporting on scans and test results may change.researchIndex: Hungarians complain about public healthcare, nearly nine in ten would back AIhealth policyPublic support could give AI rollout a social base, while dissatisfaction with care adds urgency to improvement.hospital managerPatient openness could make it easier to communicate rollout pilots.incidentYouTube AI doctors give older viewers dangerous health advice, Hungarian report saysdoctorAsk older patients about advice they picked up from AI-generated videos, and steer them to credible sources.health policyOlder people are a key target group for public information against AI-generated health misinformation.clinical deploymentWłocławek hospital to introduce AI in diagnostic imaging, with PLN 14.5m for digitisation and ransomware protectionhospital managerCybersecurity protection has to be planned alongside any AI deployment.regulationAI literacy in healthcare: why it mattershospital managerTraining in AI literacy across the institution is worth planning.caseSzeged conference presents health-industry applications of supercomputing and AIhealth ITHungarian HPC–AI capacity offers an opportunity for research and development cooperation.clinical deploymentHungarian radiologist uses AI-based diagnostics in daily patient caredoctorHungarian radiologists already have day-to-day practical experience with AI diagnostics.
More key stories of the week
insurance · United States
AI now decides access to care in six US states, and patients are being harmed
An AI pilot programme in six US states aims to curb 'wasteful' procedures, and reports say patients are paying the price.
doctorThe reasons given for prior-authorisation denials and the option to appeal should be documented.hospital managerIn the affected states, appeals and administrative capacity should be expanded.health policyPatient harm linked to the programme may prompt calls for regulatory intervention.insurance · AsiaChina to reimburse AI-assisted diagnoses through national health insurance
From 1 April 2026, China is the first country to reimburse AI-assisted diagnoses through national health insurance; the report cites 837 leading hospitals.
health policyThe reimbursement model is an example of how to finance AI-assisted diagnostics.hospital managerReimbursement shows the economic incentive for using AI.regulation · United StatesFDA's predetermined change control plan (PCCP) may play a growing role for AI devices
Industry summaries say the FDA's PCCP tool can speed up updates to AI-based devices by specifying in advance which modifications are authorized. According to a legal analysis, the FDA may give…
medtech companyScheduled, pre-approved frameworks for model updates should already be planned at the time of submission.health ITVendors should be asked for the change limits set out in their PCCP, so it is clear what can be updated without separate authorization.clinical deployment · United StatesAI scribe rollouts can ease burnout, but clinicians must be won over to use them
Mass General Brigham's AI scribe pilot significantly reduced burnout in early results, but the harder part was getting physicians to actually use it. Experience at Houston Methodist, Penn Medicine…
doctorBurnout relief depends on regular use, so ask for training and a feedback channel at rollout.hospital managerAfter the pilot, track utilization metrics and the rollout model; buying the tool is not enough.regulation · United StatesRegulating AI therapists: California lawmakers and the limits of unauthorized-practice rules
California lawmakers are trying to set limits on AI "therapists", while legal analyses say unauthorized-practice rules are of limited use for regulating mental health AI. According to the WSJ, the…
doctorPsychiatrists and therapists should track which tools are classified as therapeutic activity.health policyA ban on unauthorized practice does not on its own cover mental health chatbots, so a targeted rule is needed.health ITCompanies building mental health chatbots must expect requirements that differ from state to state.case · United StatesMore people ask AI about health problems, while trust and data quality remain open questions
Consumer Reports examines whether people can trust AI with health questions, while other reports say millions of Americans ask it for medical help but bad data stands in the way. STAT writes of an…
doctorIt is worth asking at appointments whether the patient has used AI, and helping them understand its limits.insurance · United StatesPrior authorization data goes public as health systems test real-time checks through Epic
Insurers' 2025 prior authorization data is now public, and the CMS Interoperability and Prior Authorization final rule gives stakeholders guidance. Health systems are testing real-time prior…
hospital managerHospital leaders can use the public authorization data in negotiations with insurers and when redesigning workflows.health ITThe CMS rule and real-time checks are reshaping priorities for integrating EHRs with payers.health policyThe rule's practical impact can now be measured from the public data.drug development · AsiaTCS launches agentic AI platform for drug development and clinical trials
TCS has launched an agentic AI platform to transform drug development and, under the name AgentHub, is bringing AI workers into clinical trials. The company is building the platform in India to speed…
pharmaceuticalsAgentic automation options should be mapped with trial sponsors and CRO partners.insurance · United StatesR1 to acquire Humata Health to automate prior authorization
R1 announced on August 18 that it will acquire Humata Health, adding AI-powered prior authorization automation and payer-provider collaboration to its Phare OS platform.
hospital managerR1 customers need to agree on the rollout timing of the new features and their effect on existing revenue cycle processes.health ITConsolidation toward prior authorization automation is gaining pace in the revenue cycle software market.tool · United StatesOracle Health adds coding and chart review tools to its clinical AI agent
Oracle Health is adding new coding and chart review capabilities to its clinical AI agent. Seema Verma spoke about rebuilding the EHR for the AI era.
doctorThe agent can take over coding and chart review tasks; accuracy should be checked before use.health ITOracle Health customers need to assess the deployment requirements and integration of the new agent.hospital managerThe cost and revenue effects of coding automation are best measured in a pilot.drug development · globalClinical evidence for AI drug discovery remains 'disappointingly limited' after a decade
One article says AI-based drug discovery still has little clinical evidence after a decade. Another analysis calls for a tougher test for AI models.
pharmaceuticalsFor AI-based projects, clinical-level metrics should be specified to justify the investment.clinical deployment · United KingdomHarrogate becomes first NHS trust to use AI to help breast cancer patients start treatment sooner
Harrogate and District NHS Foundation Trust is the first to introduce AI technology that helps breast cancer patients start treatment earlier. The news was published on 18 August.
doctorOther trusts may consider similar rollouts to speed up the oncology pathway.hospital managerThe Harrogate example is a model project for cutting the time from diagnosis to treatment.In brief
toolAbridge expands AI decision support beyond scribing to more cliniciansdoctorIf a scribing tool also offers decision suggestions, clinicians must check separately where documentation ends and suggestion begins.hospital managerExisting Abridge customers need to clarify clinical accountability and governance for decision support.health ITThe added functionality also raises the question of the tool's regulatory classification.regulationEU AI Act high-risk deadline pushed back 16 months; Health Canada clarifies status of AI scribesmedtech companyClassification under Annex I and Annex III must be re-planned around the new deadlines.health ITScribe products require a separate legal classification review.clinical deployment$5M grant to fund AI training for rural nurseshealth policyBuilding AI skills in rural care is also being supported at the federal level.researchReviews examine mental health risks of AI chatbots, including amplified delusionsdoctorPsychiatrists should ask patients about their chatbot use, especially when delusional or manic symptoms are present.health policyThe evidence supports setting safety requirements for chatbots.regulationRadiology Partners unit Mosaic petitions FDA on imaging AI regulationdoctorRadiologists should follow the petition, as rules on oversight of imaging AI may affect everyday use of the tools.medtech companyThe petition may affect the regulatory environment for imaging AI manufacturers.insuranceUnitedHealth: patients win 90% of appealed AI denialsdoctorAppealing denials can pay off, so it is worth doing routinely.hospital managerThe appeal rate is an important tool for protecting revenue.clinical deployment125 new NHS scanners and AI to enable one million extra tests a yearhealth policyExpanded equipment and AI together are a tool for cutting diagnostic waiting lists.doctorGreater imaging capacity means more examinations and a heavier reporting workload.researchMayo Clinic AI detects pancreatic cancer signs on CT scans originally read as normaldoctorThe tool is not yet ready for clinical use, but it shows the potential for early detection.medtech companyProspective validation is needed before market entry.toolOpenAI expands ChatGPT Health for consumersdoctorPatients increasingly arrive at appointments with questions and interpretations prepared with AI.health policyThe question of regulating consumer health chatbots is gaining weight.researchTufts–Medable: AI agents could deliver $21m net benefit per programme in clinical trialspharmaceuticalsFor the business case, the figure should be recalculated for the company's own programme structure.caseHealthcare IT private equity deal counts on pace to surpass 2021 record in first half of 2026health ITThe wave of deals could bring consolidation and acquisitions to the market.researchPolyU team develops TRUECAM framework for more reliable pathology AI in cancer diagnosisdoctorFor pathologists, frameworks that measure reliability can help in assessing AI-generated findings.drug developmentBristol Myers Squibb partners with Chai Discovery on AI antibody designpharmaceuticalsLarge companies increasingly rely on external AI biotech partners for antibody design.clinical deploymentAI helps shorten diagnosis of rare diseasesdoctorWhen symptoms point to a rare disease, an AI-based recognition tool can support differential diagnosis.researchRural hospitals are slow to adopt AI, and patients are waryhospital managerRural institutions need to treat building patient trust as part of the rollout.drug developmentNovo Nordisk–AWS partnership and Takeda bet add to AI drug discovery investmentpharmaceuticalsCloud AI partnerships point to faster research and outsourced infrastructure.clinical deploymentNHS trust launches narrow clinical AI safety check built into its patient record systemdoctorA narrow check embedded in the system may cause fewer workflow interruptions than a separate chatbot.health ITAn example of rolling out a narrow AI function embedded in the electronic health record (EHR).clinical deploymentKent hospital uses AI tool to detect infections earlydoctorAn early infection alert could bring forward the timing of intervention.clinical deploymentKT and Seoul National University Hospital to build joint medical AI platformhospital managerA partnership between a telecoms company and a hospital is a new model for platform development.clinical deploymentSheba Medical Center becomes OpenAI's first international hospital partnerhospital managerAn example of a direct partnership between a hospital and an AI developer.toolHappy Health raises $75M for AI-powered home care, starting with sleep apneadoctorHome-based sleep apnea care opens up new opportunities for providers.toolCraif raises $33M Series D for urine-based AI cancer testmedtech companyA new player in urine-based AI cancer screening has appeared on the U.S. market.caseDebate grows over physicians' role in the age of AIdoctorThe positions of professional bodies may shape the rules on liability and the division of tasks.researchEuropean projects aim to speed up the use of AI in hospital oncologydoctorThe European oncology projects point to the direction of later clinical use.caseAI helps detect health crises earlier in the German healthcare systemhealth policyThe epidemiological situation picture can be supported with AI.researchWearable sensors combined with AI enable non-invasive blood pressure monitoringdoctorContinuous blood pressure monitoring may support follow-up at home.toolSamsung develops AI models for continuous health monitoring on wearablesmedtech companyConsumer wearables are moving toward continuous health monitoring.drug developmentFilament Health signs letter of intent with Darmiyan to explore BrainSee AI in early Alzheimer's trial designpharmaceuticalsAn authorized AI tool could also play a role in selecting patients for trials.clinical deploymentAI expands rheumatic heart disease detection worldwidedoctorIn screening, AI can be useful even in areas where specialists are scarce.regulationGap between state and territorial AI policies and public health practicehealth policyState AI rules need to be aligned with public health practice.drug developmentSandboxAQ launches AI tool to cut cost of early-stage drug discoverypharmaceuticalsA new AI tool is available for early-stage research.researchDartmouth study: AI helps doctors sound more empathetic but overdiagnoses and overtreatsdoctorAI suggestions are best used for communication, while diagnostic and treatment suggestions should be treated critically.drug developmentMRC-funded £20m Cambridge hub to use AI to develop new medicinespharmaceuticalsThe academic AI hub could offer opportunities for collaboration.drug developmentEvaxion adds AI-designed off-the-shelf glioblastoma vaccine EVX-05 to pipeline, drops EVX-03pharmaceuticalsThe company is focusing on AI-designed candidates, and the pipeline reshuffle is a signal to the market.researchAI model predicts gene mutations and biomarkers from tissue images across 32 cancer typesdoctorBiomarkers estimated from histology images can complement molecular testing but do not replace it.toolAnt Group upgrades its doctor app with AIdoctorThe medical assistant app is built into hospital workflows in China.toolDCAI and Corti launch sovereign AI control layer on Denmark's Gefion supercomputerhealth ITThe need for data sovereignty is a new consideration for clinical AI infrastructure.clinical deploymentArnhem hospital uses AI to personalise depression treatmentdoctorData-based support for choosing antidepressants may reduce trial and error.licensingCureMetrix receives FDA 510(k) clearance for cmAngio 2.0doctorA new cleared AI tool is available in imaging diagnostics.medtech companyClearance of a version update points to a direction of rapid iteration.regulationAI-driven healthcare fraud and False Claims Act liabilityhospital managerAI-based coding and billing require compliance review.insuranceInsurers show growing interest in excluding AI risk from coveragehospital managerIt is worth checking liability insurance terms for exclusions covering AI use.