The Focus Shifts From the Tool, to the People Using It

On 17 September the Professional Standards Authority (PSA), ten health and social care professional regulators and the Accredited Registers Collaborative, which represents 28 accredited registers, issued a joint statement committing to develop high-level principles for how AI is regulated in professional practice across the UK. The stated aim is greater consistency between professions. It follows the recommendations of the National Commission into the Regulation of AI in Healthcare, covered in last week's issue. No publication date for the principles has been given.

Regional systems are building their own governance capacity at the same time. Liverpool City Region announced a 15-member AI Task Force on 16 September, chaired by Dr Nicola Hodson, formerly chief executive of IBM UK & Ireland. Kate Warriner, Chief Transformation and Digital Officer at Alder Hey Children's NHS Foundation Trust and chair of the North West Digital Skills Development Network, is a member. The Task Force's remit is to advise on AI development and deployment to improve public services for the region's 1.6 million residents.

PSA chair Caroline Corby said confidence in AI among the public and professionals rests on appropriate safeguards, accountability and consistency, and ICT&health reports the statement naming inaccurate information, bias, privacy concerns and overreliance among the risks. Our reading: the direction of travel is towards individual professionals being asked how they checked an AI output, so a local policy that only covers procurement will leave a gap. The principles are not written yet, which makes this a direction of travel rather than a requirement.

A Delphi consensus study in the Journal of Robotic Surgery, published on 29 August and led by co-first authors Nader Francis and Taner Shakir, proposes a national multi-specialty robotic surgery training curriculum for UK surgical residents. Across four rounds between September and December 2025, 25 respondents completed the final round and 22 of 26 statements (84%) reached the 70% consensus threshold. It appeared just before this issue's window, and we include it because its specifics are useful to any trust building a training pathway.

The panel endorsed a three-tier framework of device, basic skills and procedural training. Its specific thresholds include at least 20 bedside-assistance cases before console training (77% agreement), 10 to 30 simulator hours, and an 80% virtual-reality competency threshold (84%). Competency assessment would be built into the Annual Review of Competence Progression (84%), and all respondents agreed that deaneries and NHS funding, not trainees, should cover the costs.

The panel names access to robotic platforms and simulation as both the main barrier and the main enabler, which echoes the Royal College of Surgeons' warning that rollout is a “postcode lottery”. RCS England's guidance of 9 December 2025 already expects trusts to set up Robotic Surgery Governance Groups that approve independent practice and review procedural data, and says residents should get the same structured training opportunities as consultants.

A University of Toronto study, published in JAMIA Open in July and publicised on 15 September, found that AI scribes cut the time physicians spend documenting a visit by 69.1%. Nine primary care physicians each conducted four simulated encounters, two with a scribe and two without. Documentation took 36.3% of the encounter without a scribe and 11.2% with one. Lead author LaShawn Murray also reported: “When the scribes were in use, we saw the adoption of scratch pads for note taking to help physicians remember to ask follow-up questions.” The encounters were simulated and the study measured note-writing only, not referrals, billing or prescribing.

In Wales, Digital Health and Care Wales has opened a framework for ambient voice technology, reported by HTN as worth £10 million (PublicTechnology reported the deal at £12 million). It is due to run from December 2026 to December 2029, extendable to 2034, and covers ambient capture, transcription, generative AI summarisation, suggested clinical coding, template population and GP system integration. Tender submissions were due on 16 September and an award is expected around 30 November.

Murray's own questions point at the gap: whether patients are comfortable being recorded, and who can access the recordings. The Welsh framework's award criteria include AI performance, safety, clinical functionality and regulatory compliance, and HTN notes that the Health Services Safety Investigations Body is investigating AVT safety. In England, suppliers join NHS England's self-certified registry, which requires MHRA Class 1 medical device status and a current DTAC assessment.

On 11 September the NHS Race and Health Observatory published Towards Trustworthy, Equitable AI in the NHS, drawing on a July roundtable of 13 organisations including NHS England, NICE, the MHRA, the Patients Association, Imperial College London and the University of Glasgow. The Pharmaceutical Journal reports that it calls for anti-racism requirements to be mandated in national AI standards as procurement conditions, so that vendors supply ethnicity-disaggregated performance figures.

For NHS buyers the paper translates into four asks: monitor performance by ethnicity, tell patients when AI supports a clinical decision, publish equity-focused outcomes, and set up clear exit procedures for systems that turn out to be biased. It also recommends involving racially minoritised communities in procurement, evaluation and implementation, rather than bringing patients in after a system has already been designed.

The roundtable identified no existing mechanism to withdraw an AI tool that is causing racial harm from NHS systems. The Observatory warns that biased systems can “entrench, amplify and scale bias across every patient-facing decision”. Patients Association chief executive Rachel Power: “Patients should not be brought in once an AI system has already been designed.”