Issue 14
Viewpoint
regulation · European Union
All 53 member states of WHO's European Region unanimously adopt roadmap for AI in health care
On 2 November, the 53 member states of the WHO European Region unanimously adopted the region's first action roadmap on artificial intelligence in health care. Hungarian and Croatian analyses highlight the tasks for 2026–2033 and the governance questions.
Who is affected and how
- health policy
- National AI strategies and governance frameworks will need to be aligned with the roadmap, so it is worth mapping the gaps now.
- hospital manager
- Institutional AI governance (accountability, validation, data management) will likely be measured against this framework.
- health IT
- Procurement and compliance requirements may tighten as they are derived from the roadmap.
case · United States
Lawsuit: woman reduced epilepsy medication on ChatGPT's medical advice and died
The parents have sued OpenAI, alleging that a 31-year-old software engineer reduced her epilepsy medication dose on ChatGPT's advice and suffered a fatal seizure. Hungarian-language outlets also…
doctorPatients should be explicitly advised not to change medication doses based on AI advice, and asked about their AI use.health policyThe case sharpens questions of liability and regulation around consumer AI health advice.incident · United StatesPhishing attack on AI firm Xsolis exposes data tied to 1.4 million patients; nine health systems report impact
A phishing attack on healthcare AI company Xsolis exposed data linked to 1.4 million patients. Nine health systems reported that they were affected.
hospital managerData-handling and incident-reporting obligations of third-party AI vendors must be set out in contracts and also verified in practice.health ITPhishing protection across the supply chain and segmentation of access rights are priority risk-management tasks.insurance · United StatesElectronic Frontier Foundation criticizes Medicare's AI prior authorization pilot over delays and denials
The Electronic Frontier Foundation says the WISeR pilot hands seniors' access to care to private companies without transparency, testing or legal limits. Reports describe delays and denials in…
doctorDocumenting and appealing prior authorization denials may mean extra work.health policyTransparency, testing and appeal rights in the pilot are at the center of the policy debate.hospital managerDelays and denials from payer AI decisions create financial and care-delivery risks.tool · globalChatGPT could soon analyse Health Connect and Google Health data on Android
According to signs in the app's code, ChatGPT could soon access data stored on Android and available through Health Connect, and connect to Google Health. OpenAI says connected health data and…
doctorPatients may increasingly bring AI-analysed wearable data with them; it is worth asking about its clinical value and data source.health ITLinking consumer AI tools with health data sources raises privacy and interoperability questions.Hungary
clinical deploymentSzolnok's Hetényi Géza Hospital uses AI to support doctors after a development worth hundreds of millions of forintsdoctorIn imaging, AI support should bring faster examinations and more accurate diagnoses.hospital managerA Hungarian regional example of introducing AI alongside an equipment replacement.caseMiklós Szócska: erosion of doctors' diagnostic skills is an AI riskdoctorAlongside AI-supported diagnostics, independent clinical decision-making must be deliberately practised.health policySkill retention should also be planned into training and continuing education.caseHungarian user Vera Mérő sought weight-loss help and received life-threatening advice from an AI chatbotdoctorPatient information on weight loss and eating must take into account dangerous advice that comes from AI chatbots.health policyThe risks of AI use among the Hungarian public are a matter for public information and regulation.incidentAI-generated fake ads misuse names of Semmelweis University and BENU pharmacy chainpharmaceuticalsThe names of pharmacies and pharmaceutical retailers can be used in fake ads, so brand protection and public information need to be strengthened.health policyTackling AI scams that abuse the names of health institutions is a task for both authorities and platforms.researchSemmelweis University researcher discusses detecting type 2 diabetes risk from voicedoctorVoice-based screening could in future supplement the assessment of diabetes risk.
More key stories of the week
drug development · United Kingdom
Isomorphic Labs in talks to raise about $5 billion for AI drug discovery
Alphabet-owned Isomorphic Labs is seeking about $5 billion for AI-based drug discovery, according to Bloomberg, while Dealroom reports a $50 billion valuation. Speaking at Bloomberg Tech in London on…
pharmaceuticalsThe fundraising could intensify competition among AI-native players for partnerships and talent, and the first human trials will serve as a benchmark.research · globalStudies and professional bodies warn of harmful mental health and medical advice from AI chatbots
New studies and psychologists say AI chatbots can pose serious mental health risks, and older people are advised to be cautious about health advice from them. The UK's NHS Alliance welcomed the…
doctorIt is worth discussing separately with patients which AI tools they use for their mental health problems.health policyThe need for safeguards on the use of consumer chatbots for mental health is growing.research · United StatesAnthropic expands into life sciences: Johns Hopkins technology, drug discovery and clinical trial design
Johns Hopkins' AstroPath tumor-mapping technology has been selected for an Anthropic AI project, and the company has written about using Claude in drug discovery and clinical trials. Hungarian media…
pharmaceuticalsAI support for trial design, document review and molecule search may speed up development steps, but it must be validated.doctorAI-based mapping of tumor tissue points to a new toolkit for pathology and oncology research.drug development · globalRoche partners with Chai Discovery to deploy Chai-3 model across its pipelines
Roche and Chai Discovery announced on 5 November that the Chai-3 model will be deployed across Roche's global therapeutic pipelines. The deal extends Chai's run of AI drug discovery agreements.
pharmaceuticalsAnother example of large companies integrating external AI molecule-design platforms into their in-house pipelines.case · United StatesWhere the human-AI workflow in health care breaks down: hospitals should examine their tech stacks first
Analyses say healthcare AI struggles with workflows, system connectivity and standardization. They recommend that, before buying more AI tools, hospitals assess their existing technology stack.
hospital managerBefore procurement, integrations and data quality need to be reviewed, because the value of an AI tool depends on them.health ITRecord integration and terminology standardization are prerequisites for diagnostic AI.incident · United StatesIllinois hospitals drill for AI-driven cyberattack as security leaders outline AI agent risks
Illinois hospitals held a drill against an AI-driven cyberattack on 5 November. Security leaders at Erlanger, Atlantic Health and Lexington Health listed six risks tied to AI agents, such as forged…
hospital managerAI agents' access must be managed as strictly as that of human users, and drills must be built into plans.health ITAgent identity management, permissions and vendor access should be brought under a separate control framework.licensing · United StatesFDA clears Neurotronics' AI-based sleep analysis software
Neurotronics received FDA clearance for its AI-based sleep analysis software on 4 November. MobiHealthNews reported the decision.
doctorAutomating the analysis used in sleep diagnostics may partly replace manual scoring, but clinical validation is still required.medtech companyAnother 510(k)-type reference for the regulatory pathway of AI-based software products.clinical deployment · United KingdomNHS trials AI phone assistant for long-term conditions; Barts adds AI to breast cancer MDTs; Palantir tools underused
The NHS is trialling an AI phone assistant to monitor patients with long-term conditions, and Barts Health is introducing AI decision support in breast cancer multidisciplinary team meetings. A…
doctorMultidisciplinary team decision support and remote monitoring are being built into everyday care pathways.hospital managerUnused tools show that procurement alone is not enough; the rollout plan and actual use must be measured too.health ITActual use and value delivered by vendor platforms must be demonstrated.drug development · globalDrugmaker CEOs doubt AI's promise to cure disease as Sanofi halts a Nurix-partnered program
According to The Wall Street Journal, drugmaker CEOs are not buying Silicon Valley's AI promises, and some biotech startups are deliberately skipping AI in discovery. Nurix said Sanofi has…
pharmaceuticalsAI-derived candidates must show clinical evidence; promises do not automatically turn into pipeline value.tool · United StatesAthenahealth launches tool to let patients ask Anthropic's Claude about their health records
Athenahealth launched a tool on November 2 that lets patients query their health records through Anthropic's Claude. The integration connects patient-portal data with the consumer AI assistant.
doctorPatients may arrive with lab results and questions interpreted by an AI assistant, so physicians should be prepared to explain them.health ITConnecting EHR data to external AI assistants creates access, audit-logging and consent-management tasks.regulation · United StatesAI vendors and health systems may share liability with physicians for clinical AI, AMA says
Legal experts say that, alongside developers and physicians, the health care organizations that deploy clinical AI may also be liable for it. Modern Healthcare also covered the legal risks facing…
hospital managerContracts and deployment processes must set out who bears responsibility when AI tools fail.doctorFollowing or overriding an AI recommendation should be documented, because liability is shared.regulation · United StatesStakeholders offer competing views on how FDA should regulate generative AI in medtech
On November 3, industry stakeholders set out differing positions on how the FDA should regulate generative AI in medical technology. Five views show the main lines of the debate.
medtech companyApproval strategies for generative AI products must be adjusted as the FDA's approach evolves.health policyIn shaping the regulatory framework, the challenge is balancing the differing interests of stakeholders.In brief
researchMicrosoft's MAI-DxO diagnostic AI agent diagnoses more than twice as accurately as doctorsdoctorDiagnostic AI agents can serve as a second opinion, but the evaluation of their performance applies to controlled conditions.pharmaceuticalsMedical AI agents may reshape diagnostic careers and costs.researchAI-assisted ultrasound speeds interpretation, flags urgent cases and supports inexperienced staff in cardiac scansdoctorAI-assisted ultrasound combined with remote expert review can extend screening, but medical supervision remains essential.clinical deploymentANA launches $5M AI training for rural nurses as VCU starts AI and cybersecurity coursehospital managerBuilding nurses' AI competence is a precondition for rollouts.clinical deploymentMayo Clinic Platform launches EHR-integrated clinical decision support engine across 35+ subspecialtiesdoctorDecision support embedded in the EHR can sit alongside specialist decisions, but checking the suggestions remains the user's responsibility.health ITIntegrating with the EHR and deploying the decision support engine is a new integration task.caseGoogle sued after claims Gemini gave suicide-related content to a teenagerdoctorWith young patients, asking about the use of AI chatbots can be part of the conversation.health policyPressure is growing for AI rules that protect minors.regulationFDA classifies digital cervical cytology slide imaging systems that use AI algorithmsmedtech companyThe classification determines the regulatory pathway and the evidence required to bring cytology AI tools to market.doctorFor pathologists and cytologists, regulated use of AI-assisted screening reads moves closer.incidentAI tool use causes privacy incidents at Washington mental health facilities and Weill Cornell radiologyhospital managerA clear internal policy is needed on which patient data may be entered into which AI tool, and compliance must be checked.doctorEntering patient data into an unapproved, public AI tool can cause a privacy incident.incidentUK's NCSC says AI-powered malware 'highly likely' to be used against critical infrastructure by 2028hospital managerCyber-resilience plans and downtime scenarios must be prepared for the pace of AI-assisted attacks.health ITDetection and response processes should be reviewed on the assumption of automated attacks.regulationSenate draft bill seeks input on health AI oversight as FDA and CMS look to work with stateshealth policyCoordinating federal and state regulation is the key issue for the coming period.medtech companyMulti-level oversight may bring differing requirements, so the legislative process must be followed.researchAI-drafted patient portal messages do little to ease clinician burnout, as work shifts elsewheredoctorWhen introducing AI scribes and message drafts, measure review time as well as typing time.hospital managerTrack the impact of rollouts with workload measurements, not vendor promises.regulationEuropean Health Data Space opens opportunities for AI, while Georgetown finds gaps in AI health equity guidancehealth policyTogether, the data space and the AI Act set the framework for data sharing and risk management.health ITInteroperability and AI compliance requirements should be built into plans now.regulationStudy of 3,011 FDA warning letters finds seven citing AI, including earliest CDER letterpharmaceuticalsThe use of AI in manufacturing must be documented and validated within the quality assurance system, as the regulator already refers to it.medtech companyCompliance preparation should include validating AI tools.researchMultimodal delirium risk model validated externally and deployable without site-specific retrainingdoctorEarly delirium risk estimation could become part of preventive screening in hospital care.health ITDeployment without retraining may reduce the integration burden.incidentAI-cloned CEOs stole $129 million from hospitals and banks, US Justice Department sayshospital managerIdentity verification for approving payments and transfers must be multi-step, including for voice and video calls.regulationAMA outlines governance framework for agentic AI in clinical caredoctorThe professional body's position offers guidance on the level of oversight needed when using autonomously acting AI systems.hospital managerIt provides a reference framework for institutional governance policies.researchIn a 400-person trial, the Rosie AI chatbot did not reduce anxiety or depression in pregnant women and new mothersdoctorIn perinatal mental health, a chatbot does not replace professional care; the evidence does not support that.health policyRandomised evidence should be required before funding digital mental health programmes.toolAI writes acne prescriptions without a doctor's visit in one US statedoctorAutonomous AI prescribing raises new questions of professional liability and patient safety.health policyThe rules on AI's authority to prescribe need to be clarified.toolAgilent completes Proscia acquisition for up to $270mmedtech companyDigital pathology is consolidating, with laboratory and diagnostics players integrating AI into their platforms.pharmaceuticalsAI-based pathology data is playing a growing role in companion diagnostics development.clinical deploymentGoogle.org gives Helium Health $2.5m for AI against maternal deaths in Nigeria; Clinics On Cloud raises $12.5mhealth policyAn example of AI support for maternal care in low-resource settings.drug developmentSK Biopharmaceuticals builds large AI dataset for drug discovery as GaiaBT ties AI to lab validationpharmaceuticalsBuilding proprietary datasets and lab validation are the two decisive elements of AI research.toolnPhase raises $30M Series D for AI-native clinical data platformpharmaceuticalsData management and statistical processing in clinical trials could be automated with AI agents.researchCzech and Swiss surveys: more people turn to AI for health advicedoctorIn the surgery, doctors can increasingly expect patients, especially young ones, to arrive with preliminary information from AI.drug developmentChemLex raises $45M and launches self-driving lab for drug discovery in SingaporepharmaceuticalsAutomated, AI-driven experimental labs may become available as a service.toolHippocratic AI unveils Polaris 6 Ultra healthcare language model at a lower pricehealth ITPricing of patient-communication AI agents is falling, but vendor claims are worth checking in your own pilot.researchSurveys: most Americans uncomfortable with AI-assisted diagnoses and insurance decisionsdoctorIn patient communication, it is worth stating the role of AI openly and stressing that a human makes the final decision.hospital managerTransparent communication about AI use is needed to maintain trust.drug developmentGoogle's AI agent can propose new drug molecules within hours, DeepMind sayspharmaceuticalsAgent-based cheminformatics could shorten the time it takes to propose molecules in early-stage research.toolOctopath AI raises £1.5M to support cancer pathologydoctorPathology AI tools offer another source of decision support in tumour diagnostics.toolVera Health raises $28.5M to expand its AI clinical search enginedoctorAI search tools for clinical questions are competing for a place in physicians' daily information routines; checking sources remains the user's job.licensingFDA grants Breakthrough Device Designation to Cleo Skin AI pathology toolmedtech companyThe designation means faster dialogue with the regulator, but market authorisation still requires clinical evidence.doctorIt is an early sign of AI support becoming available in dermatological pathology.researchStony Brook study: AI may make hospital discharge summaries more efficient to producedoctorThe time needed to write discharge summaries may fall, but final review remains with the doctor.drug developmentQUANTRO wins Austrian FFG grant to build AI drug discovery platformpharmaceuticalsA Central European AI drug discovery effort receives public funding.toolMidstream Health raises $20M for AI agents that help hospitals recover missed paymentshospital managerAI-agent automation of revenue cycle management promises a financial return, which has to be measured.licensingEvident Vascular receives FDA approval for AI-powered IVUS imaging platformdoctorInterventional cardiologists will be able to use AI-supported IVUS imaging during procedures.medtech companyThe approval shows that AI can also be authorised in invasive imaging devices.researchAI predicts survival from cardiac MRI scans better than cliniciansdoctorRisk estimation from MRI scans can be automated, which may become a new decision-support tool in cardiology.researchWeill Cornell AI system could prevent testing delays in clinical laboratoriesdoctorEarly warning of laboratory disruptions can lead to faster return of results.health ITMonitoring laboratory operations is a new area for AI applications.toolGE HealthCare adds two AI features to Vscan Air for cardiopulmonary triagedoctorAI guidance on handheld ultrasound helps speed up bedside triage of heart and lung examinations.medtech companyAdding AI features to portable ultrasound is the direction of product competition.toolHeidi Health clashes with rival GP software giant over AI plansdoctorCompetition is emerging around AI scribe features in GP software, which affects the range of tools available to choose from.health ITRivalry between system administrators and vendors may affect the availability of built-in AI features.toolLeanTaaS acquires LookDeep Health to link in-room vision AI with hospital operationshospital managerCapacity and patient monitoring data may end up on one platform, which affects both bed management and patient safety.clinical deploymentSunderland project uses AI in routine eye tests to detect early signs of diabetes and heart diseasedoctorEye imaging could become a screening tool for the early detection of other diseases.