Issue 12
Viewpoint
research · global
Microsoft's MAI-DxO in Science: twice as accurate as doctors on hard cases
Microsoft AI published a study of its MAI-DxO diagnostic system in Science on 22 October; the system diagnoses sequentially using multiple AI agents, like a medical panel. Reports say it reached 83% accuracy on simulated complex cases against 36% for human doctors, and Mustafa Suleyman presents it as a step towards medical superintelligence.
Who is affected and how
- doctor
- Results on simulated cases are not yet clinical evidence, so it is worth comparing the sequential, question-asking and test-ordering approach with your own differential diagnosis workflow.
- health IT
- The multi-agent, panel-like architecture is a new pattern for diagnostic support, and integration and validation plans will need to be prepared for it.
- health policy
- Because these systems are said to outperform doctors, clinical validation and liability frameworks need to be developed first.
regulation · United States
AHA, AMA, ATA and radiology groups respond to FDA plan for generative AI medical devices
The AHA, the AMA, ATA Action and radiology societies submitted written comments in mid-October on the FDA's regulatory plan for generative AI-enabled medical devices. ATA highlighted two-dimensional…
medtech companyRisk-based classification, PCCPs and manufacturer accountability bear directly on the authorization plans for generative AI products.doctorThe professional bodies' proposals may also shape how clinician responsibility and oversight are divided.hospital managerDiverging FDA and state rules create procurement and compliance risk for generative AI tools.health policyThe conflict between state and federal levels creates an unclear regulatory environment.approval · United KingdomNICE backs interim NHS use of AI to review prostate MRI scans
On 23 October NICE backed interim NHS use of artificial intelligence in assessing prostate MRI scans. According to press reports, the tool will act as a "second pair of eyes" alongside radiologists.
doctorRadiologists and urologists may work with AI support in MRI reporting, but the final assessment stays with them.hospital managerNHS providers need to prepare for the conditions of the interim rollout, including workflow integration.medtech companyInterim NICE backing is a reference point for AI-based imaging software entering the UK market.insurance · United StatesMedicare official says parts of AI-assisted WISeR prior authorization pilot in six states are not working
A CMS official admitted that many parts of Medicare's AI-assisted WISeR prior authorization pilot are not working in the six states involved. Ohio records show the programme delayed care, and…
doctorDocumenting authorization delays and tracking patients' care in the pilot states is a practical task.hospital managerDenials and delays caused by WISeR are worth measuring and reporting to CMS.health policyThe pilot's operational problems sharpen the debate over extending AI-based authorization.drug development · globalAstraZeneca and Oracle to build AI platform for clinical trial recruitment and safety monitoring
AstraZeneca and Oracle announced on 20 October that they will use AI and real-world patient data to speed up trial patient recruitment. The agreement also covers safety monitoring.
pharmaceuticalsAI recruitment based on real-world patient data could shorten trial start-up times and serve as a model for redesigning trial processes.doctorTrial sites should expect new enquiries arising from data-driven patient identification.health ITIntegrating real-world patient data with cloud platforms raises the infrastructure demands of clinical research.Hungary
regulationPoland to join HealthAI global regulator network, second EU member to do sohealth policyJoining the network offers regional member states a model to follow for regulatory cooperation on healthcare AI.medtech companyThe international regulatory network may bring approval practices closer together.clinical deploymentEstonian hospitals introduce AI to record doctor-patient conversationsdoctorAI recording of conversations promises to ease the documentation burden, but doctors must still inform patients and check the notes.health ITIntegrating recording with documentation systems and handling the data are implementation tasks.hospital managerThe Estonian rollout is one of the most advanced uses of AI scribes in the region.regulationCzechia prepares to present its first national AI healthcare strategyhealth policyThe neighbouring country's strategy and its level of radiology AI adoption serve as a benchmark.doctorWidespread trials of AI in radiology are becoming a professional norm in the region.caseSemmelweis University lecture covers safe use of ChatGPT for healthdoctorInforming patients about their ChatGPT use is on the agenda at Hungarian professional forums.drug developmentRichter Gedeon discusses validated use of AI and advanced process control in pharmaceutical manufacturingpharmaceuticalsDomestic manufacturers are examining AI in process control built into validated production environments, which requires compliance planning.caseKóka János on healthcare's AI turn: Chinese AI hospital, randomised evidence, empathyhealth policyIn the domestic debate, randomised evidence and empathy are arguments for setting conditions on AI adoption.hospital managerThe Chinese AI hospital is a reference point in the debate on hospital AI strategies.researchHUN-REN SZTAKI develops trustworthy AI for healthcare and industryhealth ITThe domestic research programme is a potential partnership opportunity for developing trustworthy healthcare AI.
More key stories of the week
research · United States
Mental health chatbots: RAND survey finds nearly one in five young people use them
A June RAND survey found that almost one in five US 12–21-year-olds use chatbots for mental health advice, and the APA says 77% of psychologists talk with their clients about AI. Analyses say…
doctorPsychiatrists and psychologists should ask patients about chatbot use and build it into therapeutic discussions.health policyWidespread chatbot use among young people calls for protective rules.drug development · globalLilly, Astellas, Novo Nordisk and BMS strike further AI drug discovery deals
Lilly signed a $1.1 billion deal with Hong Kong's METiS TechBio for AI-based nanoparticle development, and Genome & Company also joined Lilly's AI platform. Astellas signed a $400 million deal with…
pharmaceuticalsLarge pharma companies treat milestone-based deals with external AI platforms as part of their R&D strategy.health policyA growing number of AI-driven drug candidates sharpens questions about approval and evidence.case · United StatesAI scribes save doctors time but do not fix physician burnout, analyses say
Analyses published on 22–23 October say AI saves doctors time but does not end burnout on its own. Ambient AI scribes are described as the biggest change in clinical documentation in a generation.
doctorThe time AI scribes save does not replace the need to reduce workload caused by how work is organised.hospital managerA burnout programme cannot be limited to introducing AI scribes; it also has to remove "stupid stuff" from doctors' workloads.health ITAlongside scribing tools, back-office processes and the inbox burden also need to be reduced.tool · United StatesEpic explores AI agents for routine visits and launches Connectors to link its AI tools with third-party apps
Epic is exploring AI agents for routine visits, which in some cases would remove the need for the visit altogether. On October 21 it launched Connectors, which link its EHR with third-party…
health ITConnectors and agents affect EHR integration architecture and security review.hospital managerEpic-based institutions need to plan how agent features will be introduced and governed.doctorHandling routine visits with AI agents changes the patient pathway and where decision responsibility lies.case · United StatesNurses face new liability risks from AI; union demands right to refuse unsafe technology
Reports say nurses face new liability risks from AI use, nurses have been probed over an AI error, and a union is demanding the right to refuse unsafe technology. Kaiser Permanente nurses say AI and…
hospital managerWhen introducing AI, nurse involvement and clearer liability rules are needed.clinical deployment · United StatesUniversity of Texas System rolls out Abridge AI scribe at all eight health institutions; Australia tightens TGA oversight
The University of Texas System signed an enterprise-wide agreement with Abridge, making the AI scribe available across all eight of its health institutions. In Australia, the TGA is placing AI…
doctorAs AI scribes spread, checking the accuracy of machine-generated notes becomes routine.hospital managerSystem-wide procurement is a pattern for scaling up, while tighter oversight creates compliance requirements.medtech companyThe question of regulatory classification also affects how scribe products enter the market.licensing · United StatesFDA clears MRI-based AI tool to assess breast cancer recurrence risk
The FDA cleared an MRI-based AI tool in mid-October for assessing the risk of breast cancer recurrence.
doctorOncologists and radiologists may gain additional prognostic information from MRI to inform treatment decisions.medtech companyThe clearance reinforces the regulatory pathway for prognostic imaging AI.insurance · United StatesPatients and experts oppose AI-based insurer customer service and prior authorization
A patient who has had cancer for 22 years has written an open letter asking insurance CEOs not to use AI customer service agents, and an opinion piece argues AI must not automate prior authorization…
health policyDiffering state rules on insurer AI point to a need for further regulation.doctorA documented appeals practice is needed against AI-based denials.drug development · United StatesOpenAI acquires Orchestra Bio, entering drug development
OpenAI acquired Orchestra Bio on 20 October, according to an Axios exclusive; the company builds AI agents to support drug development. The move is intended to speed up drug development.
pharmaceuticalsLarge AI companies are entering the market for drug-development agent platforms directly, creating both partnership and competition.health policyRegulatory questions about AI-developed drug candidates become more pressing.drug development · United StatesRadical Numerics says first AI-designed phage therapy has reached a human patient
California-based Radical Numerics announced on 20 October that the first AI-designed phage therapy has been administered to a human patient, and launched its Omnii-1 biological AI model. Reports say…
pharmaceuticalsThe first use of an AI-designed biological therapy in a human raises new regulatory and manufacturing questions.doctorThe clinical evidence for AI-designed phages remains open and cannot currently be treated as proven.research · United StatesSurveys: most doctors use AI, two thirds a chatbot for medical questions
A Harvard study found that two thirds of doctors use a new AI chatbot for medical questions, with about 52% of queries concerning clinical decision-making, mainly treatment and disease management…
doctorUsing AI for clinical decision support is common, so doctors need their own verification practices.hospital managerUnmanaged individual use calls for institutional policies and approved tools.health policyThe high usage rate makes validation and data protection rules more urgent.regulation · United KingdomMHRA promises closer FDA cooperation on AI medical devices at AdvaMed; Japan's PMDA shares principles
At AdvaMed 2026, the MHRA presented the next phase of medtech and AI regulation, which includes closer cooperation with the FDA and more flexible rules. Japan's PMDA published principles for…
medtech companyCloser FDA–MHRA cooperation and the PMDA principles may help companies plan multi-market clearance for AI devices.health policyRegulatory convergence makes it possible to align approval practices.In brief
clinical deploymentNHS England expands Copilot to 500,000 staff as NHS Wales launches training campaignhospital managerThe large-scale Copilot rollout offers a model for training and governance.health ITA rollout to half a million users brings data protection and licensing tasks.insuranceAI against claim denials and prior authorization: MEDITECH–Lifemed AI and Medical Economicshospital managerAI-supported appeals can be a tool for lowering the denial rate.doctorSmall and rural practices need to pay particular attention to not falling behind in authorization automation.health ITIntegrating customer-side tools to counter payer-side AI is a new area of deployment.researchAI-based survival prediction could spare some children high-dose brain radiationdoctorPaediatric oncologists may gain a new prognostic tool for personalising radiotherapy dose, after clinical validation.researchAI estimates cancer patients' survival chances and biological age from a single selfiedoctorThe results are at research stage; clinical applicability and privacy risks must be validated.health policyA face photo as health biometric data requires data-protection regulation.regulationAbu Dhabi health department signs agreements and publishes AI clinician frameworkhealth policyThe AI clinician framework is an example of regulatory oversight of research into AI medical systems.pharmaceuticalsThe AI drug discovery agreements open up new collaboration opportunities in the Gulf region.researchAI in pathology and dermatology: lung cancer biomarkers, Lunit–Summit, Mayo ClinicdoctorPathologists and dermatologists can evaluate AI as support for biomarker and biopsy decisions.pharmaceuticalsAI pathology may also play a role in selecting patients for trials and in drug development.drug developmentAnthropic, Google, ByteDance and Tencent turn to AI drug discovery; Citrini tests ClaudepharmaceuticalsThe entry of technology companies calls for rethinking research partners and supply-chain dependencies.caseOfficials and experts warn against relying on AI for medical advicedoctorDoctors should routinely ask about, and clarify, information patients bring from chatbots during visits.toolSouth African-founded Healthleap raises $38m for AI that flags at-risk hospital patientshospital managerThe market for AI platforms that flag hospital patient risk is attracting capital, which means more solutions to buy.health ITEHR integration of systems that scan hospital data and flag risk is the next deployment area.researchStudy: AI chatbots recognize unnecessary medical tests but falter where doctors disagreedoctorA chatbot's flag can help reduce unnecessary tests, but it should not be relied on in contested cases.researchRadiologists' work likely to be reshaped rather than replaced by AI; AI-generated X-rays hard to spotdoctorRadiologists should prepare for work embedded in AI workflows and for the risk of failing to recognise falsified images.hospital managerVerifying the authenticity of imaging data is becoming an integrity issue.drug developmentIsomorphic Labs raises $2.1 billion; Warp Sciences launches with $215 million; Reflector Bio emerges from stealthpharmaceuticalsCapital for AI-first biotech companies is growing, which means both partnership opportunities and competition.health policyAs AI-designed candidates move toward the clinical phase, preparing for evidence requirements is timely.toolBaba raises $40m for AI-powered Medicare care navigation and launches BabaBenchhealth ITA service that uses AI to help Medicare patients find their way is a new procurement and partnership opportunity.health policyThe insurer side (CVS) appearing as an investor signals the advance of AI in the customer journey.licensingFDA clears MRI-based AI for mild cognitive impairment prognosis and Elucid's FFR-IQdoctorNeurologists and cardiologists may gain new AI-based prognostic and functional imaging tools.medtech companyThe 510(k) clearances signal a path for further categories of imaging AI.regulationEU launches SCAILED project for health AI innovation; health data space could aid pandemic warninghealth policyThe EHDS could provide the data foundation for AI-supported epidemic surveillance.health ITStandardising data connections is a precondition for a shared data space.caseOpenEvidence and Doximity settle lawsuit; Doximity integrates UpToDate into its AI tooldoctorThe source base of AI tools used for clinical questions is changing, so it is worth checking what content the tool relies on.health ITContent licences and the legal framework for AI access are becoming more important for clinical AI products.researchPenn Medicine AI model detects hidden diseases in routine CT scansdoctorRadiologists must prepare to handle unrequested, incidental findings and to inform patients.hospital managerSecondary AI analysis of routine CTs raises questions of resources and follow-up.drug developmentPharma industry bets on AI to cut costs and timelines; scaling beyond pilotspharmaceuticalsWhen scaling AI pilots, data readiness should be scored together with business value.clinical deploymentSickKids and Microsoft sign five-year AI partnership; Ajou hospital goes live with AI monitoring; CuraVox wins awardhospital managerLong-term technology partnerships and real-time monitoring offer hospitals a model for deployment.health ITLive clinical AI monitoring places operational demands on IT teams.caseHealth systems want to build more AI, but can they afford to maintain it?hospital managerThe maintenance cost of AI models should be budgeted for at the time of deployment.health ITMonitoring and updating models requires a lasting commitment of resources.regulationAI-generated evidence in drug discovery: reliability, legal risk and FDA compliancepharmaceuticalsAI-generated citations and results must be handled with documented verification before they enter submission materials.incidentSurvey: hospital cyberattacks delay tests and procedures and raise complication rateshospital managerCybersecurity is a patient-safety issue, so downtime procedures and fallback processes must be planned.health ITThe growing number of AI systems widens the attack surface.regulationKarolinska Institutet analysis asks who should decide when medical AI may be updatedmedtech companyOversight questions around updates point to regulatory solutions similar to PCCP-style plans.researchStudy warns AI triage could push high-risk emergency patients down the queuedoctorAI triage results should be checked against clinical judgement, especially where input data are biased.