Authority Has to Be Earned

The American Medical Association published a governance framework on 13 October for agentic AI systems that take on clinical functions. Its core principle is that clinical authority is earned through evidence and kept through demonstrated performance, and that authority and oversight can change as evidence, risk or the technology changes. The framework organises its thinking around four questions: what clinical function the AI performs, what shows it can perform that function and what happens if it fails, what it should be permitted to do and under what conditions, and who is responsible for the risks and outcomes. MedTech Dive reports that the framework covers uses such as ambient documentation, diagnostics and patient monitoring, and that the AMA will work with physicians and with the people who build, deploy and regulate AI on liability, licensure and regulation. The AMA page names no requirements and assigns responsibility to no party. It is the position of a US professional body, and it is neither UK regulation nor NHS policy.

MedTech Dive notes that agentic systems can complete multi-step tasks without a human in the loop, so an error could cause harm before a clinician steps in. The AMA's fourth question, who is responsible, is posed and left open. For a UK organisation the question has a practical shape: the supplier, the deploying organisation and the individual user each hold part of the answer, and the framework does not apportion it. That reading is uPull.ai's, and the source does not make it.

Guy's and St Thomas' NHS Foundation Trust reports that its surgeons have begun using Force Feedback instruments with the da Vinci 5 robotic system, supplied by Intuitive. The trust's own page says the first operation took place on 13 October, a hiatus hernia repair carried out by Professor James Gossage, consultant upper gastrointestinal surgeon. Until now the surgeon at the console relied on vision. With the new instruments the hand controls give a sense of resistance, so the surgeon can feel the pressure applied to tissue. Channel 4 News, which described it as the UK's first operation of its kind, reports that the force sensors measure push, pull and direction about 1,000 times a second. The trust says the instruments are also in use for colorectal, gynaecological and upper gastrointestinal surgery. The trust describes them as newly approved but does not name the regulator or the mark.

The trust gives the setting for the first cases: 13 surgical robots and three diagnostic robots across five hospitals, a surgical fleet that includes da Vinci, Hugo and Versius systems alongside an orthopaedic knee robot, and more than 17,000 patients who have had robotic-assisted surgery at the trust. It reports that colorectal patients go home about four days sooner on average than after open surgery. That figure describes robotic surgery in general at the trust and is not a result for the new instruments. The first patient, Nicholas Kokkinos, says he is recovering faster than expected.

The evidence offered for the new instruments is one patient's experience and a reference to promising early US results, with no study cited and no outcome figures. The trust page does not mention training for surgeons using the new instruments. Channel 4 News makes the same point through the patient, who compares owning the technology without knowing how to use it to having a Ferrari without knowing how to drive. For a trust considering the same upgrade, three questions follow: which approval covers the instruments, what evidence the supplier can show, and how the first cases are recorded and reviewed. Those questions are uPull.ai's own, and neither source answers them.

North East Ambulance Service is running an eight-week trial of ambient voice software with 38 staff, from paramedics to operational staff. The software listens to conversations between crews and patients, transcribes them and writes a summary. It runs on the microphones of the iPads that crews already carry, so no new hardware is needed, and it is being tested in live patient encounters. The service was halfway through at the time of reporting, the trial ends in November, and the service will then decide its next steps. NEAS says it wants to learn whether the tool reduces time spent on paperwork and whether it makes records more consistent and accurate. Georgina Campbell, the service's digital clinical lead, said the aim is to use the technology safely and effectively without lowering care or record keeping standards. The report names no supplier and gives no early results. We read this through Resultsense, which republishes a BBC News report dated 13 October. The BBC page could not be opened from our systems, and the NEAS press page lists no release on the trial.

The reported safeguard is that a clinician must read and edit every AI-generated note before it is filed to a patient record. The report does not say how the trial will compare the AI summaries with the clinicians' edits, so it is unclear whether the edit rate will be measured. It also does not mention patient consent or how patients are told that a recording is made. Emergency care adds its own conditions: noise, distress and urgency. Whether those conditions affect transcription accuracy is a fair question for the evaluation, and it is uPull.ai's question, not the source's.

On 14 October the MHRA published its Strategy 2030, introduced by chief executive Lawrence Tallon. For AI, the GOV.UK announcement commits the agency to a modern, adaptive regulatory framework for AI as a medical device, a new MHRA Centre for AI and Digital Health to clarify expectations for AI-driven devices and medicines, and an AI Medicines Development Action Plan in 2027. It also commits to using AI to analyse real-world data so that safety issues are flagged earlier. The executive summary lists ten strategic priorities, including enabling technology and AI, and setting the global standard for medtech and AI. The only AI date on the GOV.UK pages we read is 2027 for the medicines plan, and the AI device framework has no stated delivery date. Last week's issue reported the government's dated steps, including draft MHRA guidance by December 2026 on changes to AI enabled devices. This strategy sets the longer horizon.

HTN's report of 15 October lists further commitments that would reach NHS organisations buying and running AI devices. They are to expand Predetermined Change Control Plans so that pre-agreed model updates can be made without unnecessary delay, to explore master files for foundation models, to make unique device identifiers compulsory through new pre-market regulations so devices can be traced across their lifecycle, to strengthen post-market surveillance in real-world use, to offer conditional approval for innovative technologies through the Early Access Service for Devices, and to support legislative reform for statutory regulatory sandboxes. HTN gives no delivery dates for these. The GOV.UK announcement adds simpler Yellow Card reporting and closer work on real-world surveillance with the Health Data Research Service.

A strategy states intent, and nothing in the sources we read changes what a deploying organisation must do today. Statutory sandboxes depend on legislative reform, and most AI device commitments are undated. The practical value for a trust is in the direction: traceability of devices, controlled model updates and continuous monitoring all point towards more evidence being expected from suppliers and from deployers after go live.