The first roundup covered the headlines of the week in healthtech, but the real story is in the details and the signals they give about the direction of the field. In this second roundup, we will review the details of the Seoul declaration, the TEMPO experiment, the Penn Medicine conversational AI platform, NVIDIA and Oracle’s strategic bets, and what we can expect in the months ahead.
The Seoul Declaration: From Principles to Action
The joint declaration of the World Health Organization and the South Korean Ministry of Health is more than a symbolic gesture. Issued at the 2026 World Bio-Summit in Seoul, it is the first time that the World health Organization and a member state have jointly issued a declaration on the role of AI in health care to the international community.
The declaration emphasizes that the benefits of AI innovation must be shared fairly, not concentrating in a few countries or groups, and that AI should be developed in regulated and responsible ways, while also encouraging solidarity between nations. Minister of Health, Jeong Eun-kyung, articulated the vision of extending health services to all citizens regardless of geography or class.
The declaration is not binding, but it does set the agenda. The session “From declaration to action” brought together a number of delegates from the WHO, the OECD, the Gates Foundation and other academic institutions to discuss how to put these principles into practice, with a particular emphasis on building the foundations for an innovation agenda, responsible use of the technology, and international cooperation on sustainable AI. One key issue that the conference considered was the evolution of artificial intelligence from a tool for learning into a self-aware and intelligent agent, one capable of generating its own hypotheses and designing its own experiments, a shift that could fundamentally alter the way in which the biomedical research and development are performed.
The signal. South Korea is positioning itself as a convening power for AI in medicine. The World Health Organization director-general’s first visit to South Korea in eleven years is a diplomatic event that will have an impact on the way other countries approach AI regulation.
FDA’s TEMPO Pilot: Real-World Evidence Before Approval
- The TEMPO (Technology-Enabled Meaningful Patient Outcomes) program at the FDA represents a significant deviation from the usual pre-market review pathway. Under TEMP, some digital health devices, including those using AI, can be introduced into real-world clinical settings before completing the usual marketing authorization, so that the agency can monitor their performance in real time.
A scope. TEMPO is based on the CMS innovation center’s Access program, which aims to improve the management of chronic diseases. Tools participating in TEMP should target the chronic diseases covered by ACCESS. One early participant, Cadence’s HypertensionOS, is a software-as-a-medical-device tool for clinician-administered hypertension medication. It uses artificial intelligence for drug initiation and dose adjustment within the framework of deterministic safety checks and predefined eligibility conditions.
The open questions. As the NEJM AI has said, TEMPO has a lot of advantages but raises some important questions regarding implementation and the longer-term consequences, particularly with regard to how the data gathered in the real world will inform regulatory decisions, and what safeguards will be introduced to prevent premature deployment of under-evidenced tools. Cadence’s participation explicitly does not signify approval or clearance from the FDA, and the agency has not determined the safety and effectiveness of its tool for the use in question.
The first TEMPO cohorts will provide data on safety, performance and outcomes that will influence future regulatory approaches to clinical AI. If successful, the sandbox model could become the standard for iterative AI tools that improve over time, a category poorly served by one-off approval processes.
UK’s Parallel Track: Lifecycle Regulation and the AI Airlock
The United Kingdom is taking a complementary approach. The MHRA’s 2030 strategy sets the ambition of a global standard for safe and agile regulation of medtech and AI by 2030, including a new center for digital health and international reliance agreements with trusted regulators, including the Food and Drug Administration.
The AI Airlock. The MHRA is already testing life-cycle regulation in its AI Air Lock sandbox, which is now in phase 3. Applications opened in October 2026, and the first wave of applicants was chosen in November. The airlock is complemented by regional partners like NHS London and University Hospital of Manchester, where real-life health care can be tested.
Government support. The UK government has accepted all 44 recommendations of the independent National Commission on the Regulation of AI in Health Care, and is now advocating a more proportionate, life-cycle approach rather than a single-point assessment. A full implementation road map is expected by spring 2027.
The Transatlantic Aspect. The MHRA-FDA Pioneer Program, announced in June, establishes a new type of exchange between the two agencies. The British Medicines and Health Care Regulatory Authority is recruiting a Deputy Director of Regional Cooperation based in Washington, D.C., to deepen collaboration between the agencies on emerging regulatory issues, including on the relationship between AI and medical devices. The prospect of a future where AI medical devices can gain faster access to the market through mutual recognition—a significant shift for companies operating in a fragmented world—is a significant change for those facing the fragmentation of regulations around the world.
Penn Medicine’s Conversational AI: From Data Collection to Collaboration
The Penn Medicine Platform, published in Nature Medicine, represents a conceptual leap: medical devices that talk to patients instead of simply collecting data for periodic analysis by the clinician…
How it works. The system has three parts: software continuously capturing EEG signals from the patient’s electrodes; a HIPAA-compliant cloud environment that analyzes the data in near real time; and a smartphone app with an AI-powered conversational agent. Patients in the epilepsy observation department could ask such questions as, “Did I have any seizures last night? “, receive visual feedback, report symptoms and respond to device-prompted alerts. The device learns from patient feedback, and when a patient flags a false alarm or confirms he had a seizure, the model is improved.
The numbers. The study included 13 adults and analyzed 1,307 interactions between patients and the system, mostly about seizure events, sleep, mood, and energy. A related NeurIPS demo cited system latencies of under three minutes and more than 25 days of continuous EEG data.
Next step: The team is preparing studies to apply the platform to deep brain stimulation for Parkinson’s disease, with Lauren Hammer, M.D., leading that work. Further applications are planned for wearables that address insomnia, epilepsy and brain and vital sign monitoring. The broader vision is to build devices that automatically adjust therapy in close collaboration with the patient, an insulin pump that answers questions about food intake, or a device for controlling tremors that adjusts in response to reported symptoms.
*** The caveat. This is a proof-of-concept in a monitoring unit, not an outpatient deployment. Scaling to home use will require addressing issues of connection, privacy of data outside of clinical settings, and over-reliance on alerts generated by artificial intelligence.
NVIDIA and Doctolib: The Infrastructure Play
NVIDIA’s minority stake in Doctolib is less about the appointment-making business and more about what Doctolil represents: a massive, structured data-set of doctor-patient interactions and a distribution channel into European health care.
The strategic logic. Doctolib has 50 million users and 600,000 health care professionals in Europe, and was valued at $ 3.6 billion in its last round of financing. The investment will allow the company to accelerate its recruitment and the development of its in-house models, mainly assistants for patients and doctors, but also administrative tools. In particular, the company wants to use NVIDIA chips to power models it owns rather than use proprietary models from OpenAI or Anthropic.
Doctolib holds appointment data, information on care pathways, medical notes, medical records and prescriptions. Although the data are anonymous for use in the training of the system, the involvement of a US giant raises questions of data sovereignty, a sensitive topic in European health. The founder of the investment, Jensen Huang, said it was about launching the AI revolution in Europe, given the continent’s lag in adopting the technology.
What to watch. The patient-facing Assistant Doctolib is already used by 400,000 French patients, and a wider rollout is planned for next year. If successful, it could become the AI that directs the health information of millions of Europeans in the future.
Oracle and AstraZeneca: Closing the Research-to-Care Loop
The non-binding agreement between Oracle and AstraZeneca focuses on one of the most stubborn bottlenecks of clinical trials: the recruitment of patients.
The plan. The companies will explore the use of artificial intelligence and real-world data to help clinicians identify candidates for clinical trials earlier and more easily, generate clinical evidence more quickly, and bring insights from research into clinical practice. The first project will support research into hypertension in Abu Dhabi in collaboration with the M42-affiliated health care company.
Beyond the recruitment, the partnership is based on a learning model whereby insights from clinical research and everyday practice can be utilised for future design, safety monitoring, and patient care. A common rule of governance, shared technical components and common outcome measures are all part of the framework.
Strategic Context. AstraZeneca is aiming at over £80bn revenue in 2030, but has been hurt by the delays to trials that hampered its pipeline. Oracle brings health data, cloud, AI and clinical practice capabilities, but the agreement is explicitly non-binding, so the execution risk is high. The success of the collaboration will depend on the degree to which the governance structures actually materialize and whether health systems can trust the data-sharing arrangements.
Future Developments to Watch
The regulatory convergence or fragmentation. The FDA’s TEMPO, the UK’s AI Airlock and the Seoul declaration all point to a life-cycle, real-world evidence-driven regulatory framework, but they are developing separately. The MHRA-FDA Pioneer Program could be the first test to see if mutual recognition of artificial intelligence medical devices is possible. If it succeeds, the developer gains a faster path to two major markets. If it fails, fragmentation persists.
The maturation of conversational medical devices. Penn’s epilepsy platform is a model. Expect to see similar pilots in cardiology (implantable defibrillators), diabetes (insulin pumps), and neurology (spinal cord stimulators) within a year or so. The critical test will be whether these systems can be reliable outside controlled observation, and whether the clinicians trust them enough to reduce their own review burden.
NVIDIA’s investment in Doctolib could be the first of many. If the models built by Doctolibers prove to be effective, other European health platforms would seek similar alliances, or would be obliged to build their own artificial intelligence capabilities. The data-use agreements that DoctoLib will establish will be a model.
Clinical trials as an artificial intelligence battlefield. Oracle and AstraZeneca are not alone, but the competition is on for EHR vendors, cloud vendors and pharma companies seeking to use real-world data for trial matching. The winners will be those that not only have technical capabilities but also governance and privacy mechanisms that satisfy the regulators and the health systems.
The clearances for Welllysis’s ECG and RingConn’s pulse oximetry systems, as well as the clearances of a number of other companies, all signal a growing familiarity on the part of the regulators with the field of AI-powered diagnostics. And the next frontier will probably be continuous multi-parameter monitoring with AI interpretation, which will not only collect data but will also give us real-time actionable insights.
Stay tuned to ztec100.com for ongoing coverage of the technology, policy and investment trends shaping the future of health.
HELEN PATRICKSON