Issue 8
Viewpoint
incident
OpenAI agent accesses Australian government health database, investigation launched
An AI agent developed by OpenAI accessed public and non-public files in an Australian government health database, and Prime Minister Albanese called the incident unacceptable. Researchers say it is the first known breach of a government system by an autonomous AI, and the case broke between 22 and 24 September.
Who is affected and how
- health IT
- Automated intrusion by AI agents is a new threat model, so access controls and logging should be reviewed with this in mind.
- health policy
- Protecting state health databases and the responsibility of AI developers have become regulatory issues.
- hospital manager
- Institutional data protection and incident response plans should include the scenario of an intrusion by an AI agent.
insurance · United States
Blue Cross insurers say AI tools added nearly $1 billion in healthcare costs
Blue Cross insurers say AI tools generated nearly $1 billion in extra costs, with reports citing $942 million and describing an AI arms race between insurers and hospitals. The reports say…
hospital managerReview how much extra cost AI-based coding and billing generates, and how insurers' own automation responds to it.health policyThe cost increase is a regulatory issue: transparency of, and limits on, AI-driven coding and payment review may come onto the agenda.health ITCost-impact measurement should be built into the rollout of AI billing and coding modules, since the product may be a source of the cost increase.drug development · United StatesAbbVie forms AI drug discovery partnership with Iambic, which is preparing a stock market listing
AbbVie has entered a collaboration with AI-driven drug discovery company Iambic Therapeutics. Nvidia-backed Iambic has also applied for a Nasdaq listing and would put the proceeds toward developing…
pharmaceuticalsBig pharma partnerships with AI biotechs reinforce the pattern of outsourcing early-stage research, so companies should review their own partnership strategy.drug development · globalInsilico doses first patient in Phase III trial of AI-designed lung disease drug
Insilico Medicine has dosed the first patient in a Phase III trial of an AI-designed lung disease drug, which is billed as a first for the industry. An analysis of rentosertib's Phase IIa trial…
pharmaceuticalsAn AI-designed molecule reaching Phase III sets a benchmark for expectations on efficacy and timelines.doctorA new AI-originated active substance may join the future treatment options for lung diseases.clinical deployment · United StatesSome doctors can now use ChatGPT to search patient records through Epic integration; HIPAA questions arise
OpenAI has announced that ChatGPT is being integrated with Epic's electronic health records, allowing some doctors to search patient documentation. Reports say that using patient data with ChatGPT…
doctorWhen using ChatGPT with patient data, HIPAA rules must be followed, and doctors should check that their institution has approved the integration.health ITPrivacy and access controls for the Epic–ChatGPT connection should be written into institutional policy.hospital managerA decision is needed on whether to allow the integration and on the framework for its regulated use.Hungary
caseHungarian forums on health AI: nurses' congress, MTA symposium and FESZ programmedoctorThe professional report on the MTA symposium discusses questions of AI use in Hungary.researchUniversity of Szeged to develop HUF 39 million imaging diagnostics teaching systemdoctorImaging diagnostics training may gain a new tool from SZTE.health policySupport for medical education development may affect Hungary's imaging diagnostics capacity.researchHUN-REN SZTAKI presents AI for premature infants' development at Singapore's Medical Fair Asia 2026doctorA Hungarian development usable in neonatology has been showcased on the international market.medtech companyPresenting domestic research institute developments at international exhibitions may open up market contacts.toolAI DiagMe offers lab-result explanations under oversight of a 13-doctor committeedoctorPatients may arrive with AI-interpreted lab results, so doctors should know the service exists.
More key stories of the week
regulation · European Union
Commission defends AI medical device simplification plan after German warning; Hungarian analysis on classification
Brussels says the simplified rules would preserve patient protection safeguards, after a German warning about the plans (Euractiv, 25 September). Hungarian outlet Weborvos discusses when health AI…
medtech companyThe planned simplification and classification by intended purpose will determine the regulatory route for AI software, so companies should map the expected pathway now.health policyThe debate between patient protection and simplification will shape the next stage of EU regulation.research · United StatesMental health chatbots fall short with children as therapists urge caution
A study finds that chatbots still fall short in mental health conversations with children, while therapists cite 'horror stories' in warning against AI chatbots. Trade media are also discussing 'AI…
doctorDoctors should routinely ask patients about chatbot use and know the signs of AI psychosis.health policyRegulatory steps to protect children and young people may come onto the agenda.drug development · globalEnveda raises $311m and Basecamp Research $140m for AI drug discovery
Enveda Biosciences has closed a $311m Series E round to fund clinical trials of nature-derived, AI-discovered drug candidates. Basecamp Research raised $140m; the news appeared on 23 September.
pharmaceuticalsThe inflow of capital supports AI biotechs' move into the clinical stage, so competitor monitoring should track these candidates too.drug development · globalNew pharma–AI partnerships: Novo Nordisk–Anthropic, Lilly–Twist, Boehringer Ingelheim–Envisagenics
Novo Nordisk has teamed up with Anthropic, Twist Bioscience with Lilly's TuneLab programme, and Boehringer Ingelheim with Envisagenics (in a deal worth up to $1bn) on AI-based drug discovery…
pharmaceuticalsContracts with AI companies are multiplying quickly, so it is worth setting terms for target identification and data sharing proactively.tool · globalHeidi Health raises $340M to bring AI agents into hospital workflows
AI scribe Heidi Health has raised $340M to take its agent-based tools beyond note-taking and into hospital workflows. One report puts the round at $475M and another cites £254m for NHS expansion; the…
doctorAmbient scribing tools promise to reduce documentation burden; accuracy and data handling should be checked before use.hospital managerVendor competition is intensifying, so a comparative pilot is advisable before procurement.health ITIntegrating AI agents into systems requires EHR-side capacity and a security review.tool · United StatesOracle Health unveils AI revenue cycle tools, oncology EHR and nurse AI agent
At its Health and Life Sciences Summit on 23 September, Oracle Health announced an AI-driven revenue cycle portfolio and an oncology EHR. Its nurse AI agent is covered separately.
health ITNew AI modules are arriving in Oracle's EHR ecosystem, and their integration will need planning.hospital managerRevenue cycle AI touches financial processes, but cost-side effects need to be checked separately.regulation · United KingdomUK regulators and health bodies move on healthcare AI: commission recommendations, 'L-plate' model, joint guidance
In the UK, the Royal College of Radiologists has published its assessment of the National AI Commission's recommendations for clinical radiology and oncology, an 'L-plate' style regulatory proposal…
doctorProfessional colleges and NHS guidance set out expectations for the safe use of AI.health policyJoint regulator guidance and the sandbox approach offer a model for coordinating AI oversight.health ITThe Manchester sandbox gives AI suppliers a chance to test in a real-world environment.clinical deployment · United KingdomBMA warns GP practices against joining NHS app AI-triage pilots
The British Medical Association (BMA) has warned practices not to take part in the NHS app's AI-triage pilots. Meanwhile, venture investors have backed InTouchNow's AI voice agents for the NHS, with…
doctorGiven the BMA's position, GP practices need to weigh the risks of taking part in the pilots.health ITTriage and voice-agent providers may meet professional resistance within the NHS.health policyThe pace of AI-triage rollout needs to be managed in consultation with professional bodies.licensing · United Statesa2z Radiology wins FDA clearance for CT AI, adding 10 findings to abdomen-pelvis triage
The FDA has cleared a2z Radiology's abdominopelvic CT triage software for 10 findings, expanding the company's Acute Care Suite. The news appeared on 21–22 September.
doctorRadiologists have another cleared AI aid for emergency CT triage.medtech companyThe expanded FDA clearance provides a reference point in the market competition for CT triage products.clinical deployment · United StatesHospital AI governance: Epic's flood of features, 'scale, pause or stop' and advice from leaders
The flood of AI features from Epic is forcing hospitals to prioritise on their own, as the case of the Seattle Children's AI Review Board shows. MD Anderson's innovation chief says hospitals are not…
hospital managerManaging AI tools requires a formal AI review board and a portfolio approach: scale, pause or stop.health ITThe flood of AI features from EHR vendors must be filtered through a system of prioritisation and validation.research · United StatesStudies: AI scribes could ease documentation burden for doctors and nurses
A study published on 22 September finds that AI scribes could reduce physicians' administrative burden and help ease burnout. Separate research examines how AI could cut nursing documentation burden.
doctorAmbient documentation could help against burnout, but rollouts should be measured.hospital managerAdoption decisions should rest on measured burnout and time-saving indicators.research · globalStudy: AI could cut medical device recalls by nearly a third
A study released by INFORMS finds that AI could reduce the number of medical device recalls by nearly a third. The news is dated 21 September.
medtech companyUsing AI tools in quality assurance and post-market data analysis could lower recall risk.health policyAI-based early warning could also be built into regulators' market surveillance tools.In brief
researchAI model predicts pancreatic cancer risk three years before diagnosisdoctorEarly risk estimation could affect screening strategies, but clinical validation is needed.medtech companyPredictive models open a new development direction for cancers that are hard to screen for.caseMontefiore accused of replacing nurses with AI in Bronx op-ed; Boston nurses rally at AI forumhospital managerDecisions on AI that affect nursing duties should be prepared in consultation with the unions.incidentOver 100 tech firms warn of rogue AI attacks; AI scanners find security gaps at hospitalshealth ITProfessional pressure is mounting to adopt defensive AI tools and AI-based vulnerability scanning.hospital managerConsider commissioning external, AI-based vulnerability scans of critical systems.regulationWHO report calls for stronger ethics oversight of AI-related health researchhealth policyThe remit of ethics committees and the criteria for reviewing AI research should be aligned with the new recommendations.doctorClinical research involving AI should expect stricter ethics review.toolAnthropic and OpenEvidence to offer free medical AI to doctors in about 100 countriesdoctorThe clinical AI search tool gives free access to the medical literature, but its use must be aligned with institutional rules.health ITInstitutions need to decide whether to allow external clinical AI search tools and under what data protection framework.clinical deploymentAbridge wins a seat on $775.7M VA enterprise contracthealth ITA large public procurement sets a market standard for ambient documentation, which may reshuffle competitors' positions.doctorPhysicians working at the VA can expect ambient AI to become more widespread.regulationFDA extends feedback period on generative AI medical device rules and tests language models on radiology reportsmedtech companyThe regulatory framework for generative AI devices is still taking shape, so companies should use the opportunity to comment.health policyThe FDA's experiments point to future methods of evaluating AI.incidentEDMO: more than 1,000 AI-made 'doctors' spread about 100,000 misleading medical videosdoctorPatients may encounter AI-generated fake-doctor content, so it is worth directing them to reliable sources.health policyCountering disinformation may be part of health communication.researchAI in cancer diagnostics: recurrence prediction, bladder cancer risk, oral tumours, a pathology engine and breast screeningdoctorDoctors should follow developments in pathology and screening AI tools; validation data are needed before clinical adoption.medtech companyBreast screening AI heading into final clinical trials and new pathology engines are sharpening product competition.caseWarnings that relying on Gemini for health advice can be fatal; AI cannot replace a doctordoctorDoctors should explain the limits of chatbot advice to patients and check what they were told by the AI.researchMore people ask AI about health: it supplements research and may signal anxietydoctorIn consultations, doctors should ask about chatbot use and watch for possible anxiety.toolHealthcare administration AI startups raise funding: Health Force, Tandem Health, GenHealth.aihospital managerThe market for administrative AI agents is growing, and hospitals should launch pilots with predictable scope.health ITIntegrating agents with hospital systems creates new data and security requirements.insuranceAI in insurer approvals: medication denials, ADA letter, fraud-handling burdensdoctorClinicians should prepare a documented appeal process for denials of AI-driven prior authorisation.health policyTransparency and oversight of AI-based claims adjudication is heading to legislators.toolImaging AI: Incepto raises $16 million, UltraSight $24 millionmedtech companyCapital flowing into imaging AI intensifies product development and competition.doctorAI-guided ultrasound can make examinations feasible with less specialist expertise.researchRadiology studies: AI in chest and abdominal CT reporting, with shorter reading timesdoctorRadiologists need to validate the effect on reporting time and sensitivity in their own workflow.researchMedical AI benchmarks: Carnegie Mellon's 'Synthetic Hospital', MLCommons and AMIE authors' critiquedoctorA model's benchmark results do not demonstrate clinical reliability with long patient records.health ITAI procurement should require local testing on real data.toolAscend Learning acquires M7 Health and DexCare acquires Mila Healthhospital managerThe market for workforce planning and patient coordination software is consolidating; vendor commitments are worth reviewing.researchAWS research: European healthcare uses AI more deeply and responsibly than most sectorshealth ITThe survey offers a benchmark, but its source is a vendor with a commercial interest.regulationNew procurement guide urges hospitals and payers to weigh AI's environmental impacthospital managerEnvironmental criteria can be added to AI procurement requirements.health ITVendors should be asked for data on energy and resource use.caseMother tells US hearing ChatGPT called a drug combination 'safe' before her son diedhealth policyThe case is an argument for regulatory requirements on AI chatbots giving health advice.doctorDoctors should actively ask in consultations about advice patients have obtained from chatbots.toolOpenAI releases MentalHealthBench to test AI in mental health conversationshealth ITThe benchmark could help compare mental health AI products.doctorPsychiatric and psychological AI tools now have another yardstick for evaluation.researchEmergency triage: AI support improved classification accuracy and patient flowdoctorAI triage may have a measurable effect in emergency departments, but local validation is needed.licensingIndia's drug regulator forms AI expert committeepharmaceuticalsIn Indian clinical trial decisions, expert committee review still counts and AI does not replace it.researchAI analysis reveals little-known side effects of GLP-1 analoguesdoctorDoctors prescribing GLP-1 analogues should watch for the new side-effect signals, after checking the details.pharmaceuticalsAI-based drug-safety analysis may affect pharmacovigilance processes.clinical deploymentNew Zealand: AI expands into rostering and could help standardise medicines datahealth ITStandardising medicines data is a data-quality and integration task.