Issue 4
Viewpoint
regulation · United States
FDA publishes discussion paper on regulating generative AI-enabled medical devices
The FDA's Center for Devices and Radiological Health (CDRH) issued a discussion paper on 18 August 2026 on regulatory considerations for medical devices that use generative AI, and is seeking public feedback. The FDA's digital health chief says regulatory guidance on generative AI is in preparation.
Who is affected and how
- medtech company
- Companies should comment on the paper, because the planned guidance will set the authorization pathway for generative AI products.
- health IT
- Teams developing or procuring software with generative functions should anticipate future FDA expectations.
- health policy
- The public consultation offers a model for regulating fast-changing generative AI within device law.
- doctor
- The status and evidence base of generative tools used on the client side may become clearer as a result of the regulation.
clinical deployment · United Kingdom
London surgeons perform first AI-assisted brain tumour operation using real-time camera image analysis
London neurosurgeons have carried out the first successful AI-assisted brain tumour removal, in which real-time camera image analysis supported the surgeons (26–27 August). The patient was operated…
doctorFor neurosurgeons, real-time image analysis during surgery may become a new decision-support tool, and its use should for now be assessed as an individual case.hospital managerThe case shows the direction of clinical pilots of surgical AI; it raises infrastructure and regulatory questions.medtech companyReal-time image analysis in the operating theatre is a development direction.incident · globalMore than 100 firms warn AI-enabled cyberattacks also threaten hospitals
More than 100 companies, including OpenAI, Anthropic, Google and Microsoft, warned in an open letter of a limited window to defend against AI-assisted cyberattacks (27 August). The letter says…
hospital managerHospital cybersecurity preparedness and insurance terms need to be reviewed.health ITDefences against AI-accelerated attacks need to be strengthened sooner.health policyProtecting healthcare as critical infrastructure is a regulatory issue.clinical deployment · United StatesHospitals adopt AI, but testing and oversight lag behind
An analysis finds that hospitals are relying heavily on AI while testing and oversight fail to keep pace (30 August). Another article says doctors are divided over AI being everywhere.
hospital managerA regime for testing, monitoring and accountability before and after deploying AI tools has to be established.doctorIt is worth learning how the AI tools in use were validated and where they fail before they are built into the workflow.health ITThe lack of AI governance processes is a risk for deployments.regulation · European UnionWHO warns Europe's health AI governance gap may become irreversible
The WHO has warned that Europe's governance gap on health AI is becoming irreversible (25 August 2026). Separately, the WHO highlighted Clalit's 77-point OPTICA safety framework as a global benchmark.
health policyNational AI governance frameworks urgently need to be put in place before the systems become widespread.hospital managerThe OPTICA evaluation framework can serve as a model for deploying AI in real-world settings.health ITSafe evaluation and deployment frameworks may become an expectation.Hungary
researchSurvey: 87% of Hungarians would support the use of AI in healthcarehealth policyStrong public support provides a social basis for AI adoption, but trust must be built on transparency.hospital managerPatients' openness makes it possible to launch AI pilots, together with patient information.doctorPatients' expectations of AI must be taken into account in the consulting room as well.caseHungarian article: AI will not replace radiologists but will change their workdoctorRadiologists must prepare for a way of working built on checking and interpreting AI outputs.researchCluj research: where AI stands in drug discoverypharmaceuticalsThere is a gap between AI promises and real results; projects need realistic expectations.caseHungarian faculty works on AI working committee led by Bertalan MeskódoctorWork on AI competencies and professional guidelines for medical education is starting at institutional level.health policyThe university AI working group could shape domestic training and professional directions.clinical deploymentRobot assistant calls patients in Budapest's 16th districtdoctorSome patient-contact tasks (calls) can be automated, but patient information and trust remain open questions.health ITA domestic example of AI-based automation of patient communication.caseSemmelweis HELP: AI should not be the first "doctor" when a child is illdoctorPaediatric care must expect parents to have consulted AI chatbots beforehand.
More key stories of the week
drug development · global
AI drug discovery: Nvidia backing, Novo Nordisk–AWS partnership and Insilico results
Nvidia is backing Generate Biomedicines (25 August), Novo Nordisk and AWS have formed an AI partnership to speed up drug discovery, and Insilico Medicine has published its half-year results. One…
pharmaceuticalsLarge companies, cloud providers and AI startups are dividing up the work; partnership strategy must be aligned with this.health policyPhase III results for AI-designed candidates will shape how the sector is perceived.research · globalAI analysis of mammograms can also detect heart disease in women, study suggests
Researchers say AI analysis of routine mammograms can also flag cardiovascular disease in women (The Guardian, 27 August). Experts say breast cancer screening could therefore also point to…
doctorSecondary analysis of screening images could provide a new cardiovascular risk signal, but it requires clinical validation.hospital managerData from existing screening programmes could add value if AI analysis can be built into the workflow.medtech companySecondary risk estimation based on screening imaging is a new product area.research · globalAI cancer detection results: bladder, liver and lung nodules, plus an Estonian study
AI reportedly flagged bladder cancer up to five years before diagnosis, and a Chinese model found 15 missed liver cancer cases in a clinical trial. An Estonian report says AI detects cancer better…
doctorThe results point toward earlier detection and fewer missed cases, but the clinical trial details will decide the strength of the evidence.hospital managerJustifying pilots of AI-based screening support requires measurable sensitivity and false-positive data.case · United StatesNurses across the US protest hospital use of Palantir's AI technology
National Nurses United organised a nationwide protest involving nurses on 27 August against Palantir's role in hospitals. The nurses' union says AI interferes with clinical judgment and threatens…
hospital managerIn choosing AI vendors, resistance from staff and the community has to be taken into account.health ITThe role of data-platform providers in hospitals has become a subject of public debate.drug development · globalAI drug discovery: benchmark flaws may distort leaderboards, as CIOMS issues pharmacovigilance validation standards
An article reports that benchmark flaws may be distorting AI drug-discovery leaderboards (27 August). The CIOMS Working Group XIV has released validation standards for the use of AI in…
pharmaceuticalsVendor claims based on benchmarks should be verified independently; a validation standard is now available for pharmacovigilance AI.health policyThe CIOMS standard could serve as a basis for regulatory expectations on pharmacovigilance AI.research · United StatesBehavioral health AI: Onos Health raises $17 million, chatbot therapy rated, triage concerns
Onos Health has raised $17 million for its AI-based behavioral health platform (26 August). Psychologists rated an AI chatbot's therapy sessions on the scale used for human therapists, and California…
doctorA chatbot may suit basic therapeutic conversation but is weak at applying specialised techniques consistently, and triage delays pose a danger.health ITSafety checks are key when deploying AI triage in mental health care.health policyLawsuits and FDA consultations are pushing for regulation of mental health AI.research · globalRadiology AI: cognitive biases, risks of ruling out findings and misidentified AI images
One report says cognitive biases can mislead radiology large language models (26 August), and another urges caution when radiologists use AI to rule out suspicious findings. Other reports say…
doctorUsing AI as a rule-out tool is risky; the verification steps in reporting must be deliberately kept.hospital managerWhen deploying AI, accuracy must be measured alongside speed gains.medtech companyTesting devices for human factors and biases is becoming an expectation in their evaluation.research · United StatesSurvey: patients want to know when their care provider uses AI
US surveys find that patients expect transparency when AI is used in their healthcare (25 August 2026). Modern Healthcare also reported on the survey.
hospital managerIt is worth drawing up a patient-information rule on when and how to disclose the use of AI.doctorInforming patients about the use of AI can be part of doctor–patient communication.health policyPatient expectations support the case for AI transparency rules.research · United StatesPew: more than a third of Americans have turned to an AI chatbot for health advice
A Pew Research Center report finds that slightly more than a third of Americans have used conversational AI for medical guidance (PCMag, 26 August). Another article says AI medical advice can also…
doctorPatients increasingly arrive with information from a chatbot; this has to be handled during the visit.health policyThe scale of public chatbot use justifies regulatory and patient-information steps.regulation · European UnionJRC: AI can improve disease surveillance in Europe, but human oversight is essential
The European Commission's Joint Research Centre says AI can improve disease surveillance in Europe, but human oversight remains essential. A separate article covers the secondary use of data under…
health policySurveillance AI must be designed with human oversight; EHDS data will be opened up for research and AI purposes.health ITData management must be designed to comply with the EHDS rules on secondary use of data.drug development · United StatesFaro raises $37 million to speed up clinical trials with AI
Faro has closed a $37 million Series B round to speed up clinical trials with AI (26 August). Another article says AI could accelerate trial design and patient recruitment.
pharmaceuticalsAI built into protocol design and patient recruitment can cut the time and cost of trials.regulation · United KingdomAI scribe risks: insurer checklist, documented errors and MHRA guidance on medical device status
The Doctors Company has issued a checklist for reducing the risks of AI scribes (26 August). One article cites a case where a scribe inserted a false statement into a note, and the MHRA clarified…
doctorGenerated notes must be checked before signing, and patient information about the tool's use must be arranged.hospital managerLiability and risk management must be clarified in contracts with scribe vendors.health ITAccording to the MHRA, the line between administrative AI and a medical device must be watched in contracts and marketing as well.In brief
licensingFDA clears two AI tools: pulmonary hypertension detection from ECGs and sepsis predictiondoctorNew early-warning tools are becoming available based on routine tests (ECG) and clinical data.hospital managerDeploying sepsis prediction brings measurable outcomes as well as an alert burden.medtech companyCleared examples provide a reference for AI-based early-warning products.researchSurvey: physicians want higher pay from AI productivity gainsdoctorAI use can become an argument in pay negotiations.hospital managerHow productivity gains are shared must be decided before deployment.health policyAI-driven productivity is part of workforce retention and pay policy.licensingLimbic selected for FDA pilot to test voice AI therapy with Medicare enrolleesdoctorMental health AI tools are entering an authorised pilot with measurable outcomes.medtech companyThe FDA pilot shows the regulatory path for mental health AI products.health policyMedicare's scope also brings the question of how AI therapy is funded to the fore.toolEpic turns its EHR into an AI command center, with Ergo platform and Anthropic tie-uphealth ITAn EHR vendor's AI platform shapes the integration strategy for third-party tools.hospital managerEpic's AI roadmap underpins purchasing and deployment decisions.doctorAI and decision support built into the EHR appear in the everyday working interface.regulationFDA updates cover substantial evidence, gene therapy and AI-enabled early-phase clinical trialspharmaceuticalsThe FDA's position on the use of AI in early-phase trials should be taken into account in development plans.insuranceInsurers say AI does not deny care; senator criticises AI-based Medicare programmedoctorPhysicians should expect AI-based prior authorisation and denial processes in practice, and document the reasoning behind decisions.health policyTransparency of insurers' AI use and oversight of AI-based Medicare programmes are political issues.licensingFDA grants two breakthrough designations to generative AI radiology productsmedtech companyThe designation can mean expedited engagement with the FDA and is a reference point for radiology generative AI developers.doctorGenerative AI radiology tools may move closer to clinical use.toolSuki launches standalone AI clinical dictation tool with native Epic and MEDITECH integrationdoctorAI documentation based on dictation is available as an option, without ambient listening.health ITThe Epic and MEDITECH integration allows simpler deployment.hospital managerThe unbundled package allows more flexible procurement.clinical deploymentAI in screening: faster breast cancer testing in Harrogate, hypertension and diabetes signals from facial videodoctorExperimental facial-video methods are not yet validated and cannot be the basis of diagnostic decisions.hospital managerPathology AI can speed up the breast cancer diagnostic process.caseRural healthcare: patients wary of AI solutions as hospitals struggle to keep uphospital managerIn rural institutions, AI adoption needs to be built on patient trust and resources.health policyAcceptance of rural AI solutions and resource inequality are regulatory questions.clinical deploymentNortheast Ohio hospitals test AI to improve care and cut costshospital managerThe experience of local hospitals is a benchmark for measuring the costs and benefits of your own AI pilots.regulationAMA launches Ethical AI Use in Medicine SeriesdoctorThe series offers professional ethical guidance on the clinical use of AI.health policyThe professional body's position is a reference point in regulatory debates.toolArintra closes $25M series B for AI-driven medical codinghospital managerAutonomous coding affects the cost and error rate of the revenue cycle.health ITAnother AI coding player enters the market for revenue cycle solutions.clinical deploymentNHS trusts roll out AI technology; 90% of clinicians say they use itdoctorAI use is already widespread in British care, so guidelines and training are needed.health ITHigh usage creates deployment and governance requirements.regulationLegal concerns raised over FDA's 'competency' approach to AI regulationmedtech companyGiven the legal uncertainty, compliance strategy and regulatory communication should be reviewed from a legal perspective too.health policyThe legal basis of the regulatory approach may be contested, which could affect enforceability.researchAnalysis: FDA's AI device list holds 1,430 authorized devices for 1995–2025medtech companyThe specialty distribution shows which clinical areas are crowded or still empty in the AI device market.health policyThe registry analysis provides data for regulatory monitoring of how AI devices spread across specialties.clinical deploymentGEMINI launches AI-assisted delirium prevention trial across 13 hospitalsdoctorAn AI-assisted protocol for preventing hospital delirium is being tested, affecting physician training and data entry.hospital managerThe multi-hospital trial offers a model for AI adoption built on professional collaboration.caseDuke launches new AI specialization for nurses in clinical practicehospital managerThe training offer is a model for developing nursing staff's AI competencies.toolPhilips wins up to $33.7M ARPA-H award for AI-enabled robotic stroke caremedtech companyGovernment research funding is behind autonomous robotic interventions.doctorAccess to stroke care would be expanded through robotic AI.toolUCSF system gives confidence ratings to support brain tumour MRI analysisdoctorThe confidence rating can help decide when an AI suggestion needs particularly careful checking.medtech companySignalling uncertainty can be a distinguishing feature of AI image-analysis products.regulationIsrael's regulatory sandbox tests autonomous AI in clinical carehealth policyThe sandbox is a model for gaining regulatory experience in a controlled setting.medtech companyIt offers a testing opportunity for autonomous AI tools.clinical deploymentUPMC expands AI-native transitional care model to 18 hospitalshospital managerAI support for care after hospital discharge can be measured even in a large health system.toolMiiHealth AI closes seed round to rework patient intakehealth ITAutomating patient intake is another entrant in the market for AI in provider administration.regulationRadiologists in Brazil debate safety rules for medical AIdoctorThe professional position of radiologists is shaping national AI safety rules.health policyThe professional debate is a factor in drawing up national regulation.