Issue 5
Viewpoint
incident · United Kingdom
Healthwatch: AI scribes leave drug and diagnosis errors in medical notes
According to a report by Healthwatch, the patient watchdog for England, patients are spotting drug and diagnosis errors in AI transcripts that NHS doctors miss. Under the conditions of their exemption, the systems require clinician checks, and at least 27 different AI scribe platforms are in use.
Who is affected and how
- doctor
- An AI note must be checked item by item before signing, especially drug names and diagnoses, because the check is a condition of use.
- hospital manager
- The institution should require a checking protocol and an error-reporting channel for AI scribes, and involve patients in reviewing their notes.
- health IT
- Error rates should be tracked per AI scribe platform, and the clinician approval step documented in the system.
- health policy
- The parallel use of nearly thirty platforms calls for uniform quality and oversight expectations.
clinical deployment · United States
OpenAI connects ChatGPT to Epic data for clinicians
OpenAI announced on 1 September that ChatGPT for Healthcare is connecting to the Epic EHR system and public health data sources. According to reports, clinicians can import data such as notes in…
doctorDoctors working in an Epic environment should try the read-only data import, but must still check ChatGPT's output against the source records.hospital managerAn institution that uses Epic must decide whether to allow the integration, and under what data-handling, access and liability rules.health ITThe integration's data-flow, permission and logging issues must be assessed before it is switched on in the institution's Epic environment.medtech companyThe linking of major EHR vendors with general-purpose AI platforms may narrow the space for standalone clinical AI tools, so interoperability strategy should be reviewed.licensing · United StatesFDA pilot offers generative AI medical devices a path to patients before authorization
According to reports on September 3, the FDA is launching a pilot that offers generative AI medical devices a path to patients before authorization. Earlier in the week, it was reported that the FDA…
medtech companyCompanies developing generative AI should examine whether to take part in the pilot and prepare for post-market monitoring.health policyThe regulatory approach could offer a model for other authorities handling generative AI devices.doctorUnder the pilot, devices may reach patients before authorization, so clinicians need to know a device's validation status.licensing · European UnionVara receives world-first CE certification for autonomous AI in breast cancer screening
On September 2, Vara received the world's first CE certification for autonomous AI in breast cancer screening. The certification marks a regulatory breakthrough for independently operating diagnostic…
doctorRadiologists need to prepare for AI working independently in some screening cases.hospital managerBreast screening programs should consider procuring autonomous AI and redesigning their workflows.medtech companyAutonomous diagnostic AI now has an EU regulatory pathway, which is also a benchmark for competitors.incident · globalHospitals' AI cyber risk: joint letter, report and medical device vulnerabilities
On 27 August, OpenAI, Microsoft, Google, Amazon, Anthropic and Cloudflare warned in a joint letter that hospitals and other critical infrastructure are unprepared for AI-assisted attacks. A Black…
hospital managerHospitals adopting AI need to expand cybersecurity controls in step and review their patching and incident response processes.health ITUpdate and incident response procedures need rethinking, because AI speeds up the exploitation of vulnerabilities.medtech companyMedical device security must be designed on the assumption that flaws can be found quickly with AI.Hungary
researchSurvey: nearly half of Hungarian internet users have already asked AI a health questiondoctorPatients increasingly arrive with information and self-diagnoses from AI, which must be addressed during consultations.health policyPublic use of AI justifies health information and health literacy programmes.health ITThe data on Hungarian user trust can inform the design of Hungarian-language health AI products.clinical deploymentPoland's health ministry to bring AI to nearly 400 hospitals; platform due by December 2026doctorRadiologists and doctors working with imaging diagnostics need to prepare to use the AI platform.health policyThe national AI platform is also a point of comparison for Hungarian health policy.hospital managerHospitals need to make preparations to connect to the AI platform.researchOnly one in four Hungarians is satisfied with healthcare; 48 percent hope for faster AI diagnosishealth policyDissatisfaction together with the hope for faster AI diagnosis points to public support for digital developments.hospital managerPatient expectations point towards faster, AI-supported care.regulationAct XLIV of 2026 on health data management covers EESZT and refers to AIhealth ITDevelopers working with EESZT need to review the act's data management and access requirements.health policyThe act is part of the health data management framework, and its implementation is worth following.doctorThe rules on access to children's EESZT data may affect everyday practice.casePharmacists and EESZT: robotic systems and AI offer support, not replacementtoolOPSEYA uses AI to prepare scans and flag areas for review, EESZT-compatibledoctorThe AI pre-screening flags areas for review, but the final assessment remains with the physician.health ITEESZT compatibility is a prerequisite for integration in Hungary.caseAI could order hospital supplies instead of manual stocktakinghospital managerAutomating inventory management can improve operating efficiency, but return on investment should be measured before rollout.researchSemmelweis University workshop on machine learning in public health under HAI-SO projecthealth policyPublic health AI models are being developed to support decisions on health system optimisation.hospital managerThe project's results can be used in institutional planning.researchSemmelweis University: structured data and AI to improve dental caredoctorStructured documentation is a prerequisite for AI support in dentistry as well.health ITStructured data models can provide the basis for AI development.
More key stories of the week
insurance · United States
Bayesian Health's sepsis monitor wins Medicare new technology add-on payment
Bayesian Health's FDA-cleared AI sepsis monitor was approved for a Medicare add-on payment on 2 September. The decision is a rare example of reimbursement for an AI-based clinical tool.
hospital managerReimbursement improves the business case for deployment, so buying a sepsis monitor is worth assessing.medtech companyThe Medicare payment offers a template for the reimbursement strategy of AI tools.doctorSepsis alerts are likely to spread, so alert protocols should be clarified in advance.clinical deployment · United KingdomMicrosoft rolls out Copilot to 505,000 NHS staff
According to BuildApps, Microsoft is rolling out Copilot to 505,000 NHS staff, the largest AI software deployment in UK healthcare. Analyses say it also tests the NHS's enterprise AI readiness.
hospital managerUsing Copilot requires data protection and usage rules.health ITThe NHS experience offers lessons for large-scale enterprise AI deployment.regulation · European UnionEU AI Act deadline for medical devices reportedly moves to 2028, with two separate deadlines
The EU AI Act deadline for medical devices is reportedly moving to 2028. Analysis suggests health AI faces two deadlines eight months apart, depending on whether the product is legally a medical…
medtech companyCompliance roadmaps need to be re-planned, and companies must clarify which deadline applies to their product.health ITSoftware should be assessed for whether it qualifies as a medical device, as that determines the deadline.health policyThe postponed deadline gives a longer transition, but institutions also need to prepare.clinical deployment · United KingdomNHS to trial two-second AI ECG analysis for hidden heart disease in six hospitals
The technology, trained on 10.6 million ECGs, is being tested in six NHS hospitals ahead of national rollout. The AI can flag hidden heart disease within two seconds, which reports say could also…
doctorFor cardiologists and GPs, AI-ECG could support referral decisions.hospital managerBecause it could reduce waiting lists, the AI-ECG pilot deserves attention.health policyA six-hospital test before national rollout is a model for evidence-based scaling.case · United StatesKaiser nurses to hold one-day picket over AI-use concerns
Kaiser nurses are to hold a one-day picket over concerns about the use of AI, according to reports published on August 31, 2026. The action is the latest in nurses' unions' pushback against AI.
hospital managerInvolving employee representatives early when introducing AI tools can reduce the risk of conflict.regulation · United StatesBipartisan US bills target AI-driven insurance claim denials
Reps. Barrett and Landsman, the latter with Republican and Democratic colleagues, introduced bipartisan legislation on 1 September to stop AI from making health care decisions and protect patients…
doctorThe legislation could reduce AI-automated denials, so developments are worth following.hospital managerDenial rates and appeal burdens may change, which should be factored into revenue planning.health policyBipartisan support may signal tighter rules on insurers' use of AI.research · United KingdomStudy: AI scribes may miss patients' experiences and cause deskilling
A study published on 3 September says AI note-taking tools may change the interaction between patient and clinician and leave out important information. The authors also cite the risk of staff losing…
doctorPatients' accounts of their experiences should be deliberately recorded during the consultation, because the AI transcript may omit them.hospital managerWhen introducing AI scribes, attention is needed to preserving documentation skills and to measuring patient experience.health ITWhen evaluating an AI scribe, content completeness should be measured alongside time savings.drug development · globalOwkin signs AI drug discovery deal with Boehringer Ingelheim for cancer and immunology
On September 2, Owkin signed an agreement with Boehringer Ingelheim on AI-based drug discovery in cancer and immunology. The deal is another example of cooperation between large pharmaceutical…
pharmaceuticalsThe collaboration model shows that large companies are expanding their research pipelines with external AI platforms.research · United StatesStudies link AI chatbot use to loneliness and emotional problems among young people
One study finds that using chatbots for emotional support is associated with loneliness among students. Another says chatbots may be linked to higher emotional problems in teenagers, while RAND says…
doctorIt is worth discussing with young patients why and how often they use AI chatbots.researchStudies: AI scribes cut doctors' administrative time
Research from the University of Toronto finds that AI scribes save doctors time and may ease the administrative burden. According to a report, clinicians at OLVG spend more than 30% less time on…
doctorAn AI scribe likely cuts documentation time substantially, but time for review must be planned in.hospital managerThese figures can support an institutional business case, but return on investment should also be measured in a pilot in your own environment.regulation · United KingdomManchester NHS and MHRA launch AI testing sandbox; MHRA commission publishes findings
Manchester NHS and the MHRA launched a health innovation sandbox on 3 September to test AI and new technologies. The MHRA's commission has published its findings on AI in healthcare.
medtech companyThe sandbox gives AI tools a chance to be tested in a regulatory environment.health policyThe regulatory sandbox provides experience for refining regulation.case · United StatesUnion alleges nurses are being displaced by AI at Montefiore; nurses also protest Palantir
Crain's reports that the union at Montefiore claims nurses were let go after the introduction of AI. Another report says nurses are protesting over Palantir's healthcare technology and staffing…
hospital managerStaffing and job-role consequences should be communicated and discussed before the technology is introduced.In brief
caseExperts: AI scribes carry malpractice risk, but clinical AI may also serve as a defense assetdoctorThe physician who signs an AI-generated note is responsible for its content, so documenting the review also serves as legal protection.hospital managerBefore deploying AI tools, hospital leaders should consult their liability insurer about coverage and policy.licensingWeave Bio and Takeda deploy AI regulatory automation across 14 global programspharmaceuticalsAI automation of regulatory submission documents can speed up regulatory workflows, so a pilot is worth starting.researchStudy: AI chatbots wrongly reassure sleep apnoea patients in a third of casesdoctorChatbot reassurance can delay diagnosis, so it is worth asking patients about AI use during history-taking.researchAI model flags colorectal cancer on routine noncontrast CT scansdoctorFor radiologists, AI could flag colorectal cancer as an incidental finding on routine CT, after clinical validation.hospital managerOpportunistic AI screening could build on existing imaging capacity.regulationSan Francisco supervisors urge Kaiser to limit AI use in mental health servicesdoctorMental health professionals need to clarify which tasks they may hand over to AI in patient care.health policyLocal bodies can also put pressure on providers to limit the use of AI in mental health care.researchTransluce publishes 50,000-plus simulations testing how AI systems handle mental health crisesdoctorMental health professionals need to know how chatbots handle crises before recommending them to a patient.health ITDevelopers and integrators must test how AI chatbots handle crisis situations.researchOf 1,357 FDA-cleared AI devices, only three were tested on patientsdoctorClearance of operating-room AI tools does not mean effectiveness has been shown in patients, so clinicians should ask separately for clinical evidence.medtech companyThe lack of clinical validation risks tighter expectations from regulators and buyers.hospital managerAt procurement, hospitals should ask manufacturers for prospective patient studies.regulationEU court ruling clarifies medical device definition for software and AImedtech companyIntended-purpose wording and function should be reassessed, as they decide whether AI software qualifies as a device.health ITWellness and decision-support software warrants legal review.researchBlind spots found in ChatGPT Health's medical triagedoctorConsumer AI triage can be wrong, so patients should be educated about the limits of using it.clinical deploymentNHS app to use AI to direct patients to the right service in EnglanddoctorPatients may arrive via AI routing, which could affect GP and emergency triage.health ITAI routing in the NHS app requires integration of patient pathways.toolOpenEvidence deepens oncology focus, launches new AI model familydoctorA specialized decision-support tool is being built for oncologists; its evidence should be assessed before use.drug developmentNew AI drug discovery tie-ups and projects: Ono–Aitia, Hiro, MND, JapanFoldpharmaceuticalsThe partnerships and local data-handling arrangements offer examples for procurement and partner-selection decisions.clinical deploymentPunjab approves AI CT scanner that detects 10 diseaseshealth policyProvincial-level rollout of AI diagnostics is an example of speeding up public services.regulationGuidelines needed as mental health practices adopt AI, reports saydoctorMental health professionals need to develop their own protocols for using AI.health policyThe lack of professional guidelines justifies regulatory action.researchImaging and pathology AI research: ATTR-CM, coronary CT, breast cancer, HCC, SlideChat, endometriosisdoctorThe new studies show directions for diagnostic decision support, but validation is needed before clinical use.caseNurses face a 'visibility issue' with AI in healthcare and want it to serve themhospital managerInvolving nurses in AI adoption improves how well the tools fit workflows.caseChatGPT conversations can end up in court and are not protecteddoctorIt is worth telling patients that health information shared with a chatbot is not covered by medical confidentiality.drug developmentInsilico Medicine releases frontier AI models for drug discoverypharmaceuticalsResearch teams should evaluate the new models against their own target and molecule identification workflows.regulationUS medical bodies: ACP issues ethics paper on AI, AMA names head of new digital health and AI centerdoctorThe ACP's ethics paper sets a professional benchmark for the use of clinical AI.drug developmentAI drug discovery draws doubts as venture bets yield little and most drugs still failpharmaceuticalsWhen investing in AI, companies should ask platform vendors for evidence measured in clinical success rates.toolAI is reshaping pathology as diagnostic investment growsdoctorPathologists need to prepare for digital and AI-supported workflows.medtech companyInvestment flowing into pathology AI signals a market opportunity.case$5 million grant to train rural nurses in AI; Johns Hopkins publishes AI framework for nursing educationhospital managerRural providers may gain access to training funds for AI adoption.incidentBurnout hits cybersecurity staff defending hospitals and banks, Bloomberg reportshealth ITSecurity teams' workload and the options for automation need to be managed.hospital managerRetaining cybersecurity staff and managing their workload is an operational risk.toolAMA: what doctors want patients to know about health wearablesdoctorPatients' wearable data will show up in the clinic, so its limits of interpretation need to be explained.clinical deployment10,000 Queensland doctors use Heidi Health's AI scribedoctorAI note-taking has spread at large institutional scale, so doctors must also develop checking habits.hospital managerExamples of large-scale deployment can serve as references for procurement decisions.health ITMass deployment yields EHR-integration and data-protection patterns that are worth studying.caseRural hospitals use AI tools to counter insurers' claim denialshospital managerAI tools can be used to manage denials, but return on investment should be measured.health ITThere is a market for denial-management AI in revenue cycle management.researchAI analysis of facial video explored as a way to flag hypertension, diabetes and heart diseasedoctorRemote, contactless screening deserves attention, but clinical validation should be awaited.regulationUK launches £100mn competition for homegrown AI start-ups to improve the NHS and public serviceshealth ITBritish start-ups have a funding opportunity for NHS-related AI solutions.health policyThe competition is an incentive tool for AI-based public service improvement.caseCureety acquires Reimagine Care to advance AI-driven precision oncologyhealth ITProviders of digital oncology care are consolidating.regulationWHO forum: progress on health AI should be determined by strength of governancehealth policyEstablishing governance frameworks is a precondition for AI adoption.researchASCO: clinical AI needs more and better benchmarksdoctorThe lack of independent benchmarks calls for caution when evaluating oncology AI tools.researchAI in digital dermatoscopy may improve diagnostic performancedoctorFor dermatologists, AI decision support can improve diagnostic accuracy when it assists the physician's decision.clinical deploymentAI-powered ultrasound gives clinical decision within 30 minutes for suspected blood clot patients at Chelsea and WestminsterdoctorAI ultrasound can speed up decisions when deep vein thrombosis is suspected.hospital managerIt is an example of shortened patient pathways for rapid diagnostic pilots.regulationSouth Korea's 2027 budget plan sets aside 1.26 trillion won for regional and essential healthcare, with more medical AIhealth policyState budget priorities are shifting toward AI and care robots.medtech companyState funding for medical AI and robotics is a market opportunity.regulationGovernment backs 372 AI projects in business and healthcare sectorshealth policyState AI funding could serve as a model for funding AI in healthcare.caseMalaysia's IJN Group began AI training with its CEOhospital managerAI literacy among leaders is a precondition for adoption.caseFunding rounds: sleep apnea platform, AI patient coordination and clinical AI startuphealth ITInvestment is flowing into sleep apnea, patient coordination and clinical AI.