The Rollout Runs Ahead of the Plan

On 21 September the National Cyber Security Centre published a blog by Dave Chismon, its CTO for Architecture, titled One does not simply defend agentically. UKAuthority reported it on 28 September. Chismon argues that attackers gain from AI because their goals are technical and their success is easy to measure, while defenders work within organisational constraints such as budget approval, system replacement timelines and staff availability. He sets out three principles: technology detects and humans respond, detection must not harm operations, and responses are strictly controlled and narrowly scoped. The blog is general guidance for organisations and does not mention health.

Rotherham NHS Foundation Trust offers an example of a bounded first agent. Digital Health reported on 22 May that an AI agent from Netcall resolves routine IT queries, raises categorised service tickets and passes complex issues to IT staff with a pre populated ticket. IT help desk call volumes fell by 28%, and 41% of IT queries are now handled through self service and AI agents. A second phase will filter non urgent out of hours requests before they reach on call staff. This story is from May. We include it because UKAuthority's summary of the NCSC blog suggests starting with low risk tasks, and an IT help desk is a clear example of one.

HTN reported on 1 October that NHS England's board kept its cyber risk score at 25, the highest rated risk on its register, because of maturity concerns across the health system. Set beside the NCSC's principles, the question for any agent able to act on NHS systems is who reviews each action and how quickly a human can step in. The Rotherham tool routes complex issues to staff, which is one way of writing that answer down. The link between these two sources is uPull.ai's reading, and neither source makes it directly.

ITV Meridian reported on 4 September that surgeons at Maidstone and Tunbridge Wells NHS Trust in Kent reached full theatre schedules in gynaecological cancer, general surgery and urology on the da Vinci Xi system faster than any hospital in the UK and Ireland over the previous seven years. The trust received its first system in September 2024 and has since carried out more than 1,000 operations with the robots.

Daniel Lawes, a consultant colorectal surgeon at the trust, links the platform to training, saying it lets the trust offer the best training and development opportunities to its staff and to future surgeons. The report gives volumes and speed of adoption. It gives no outcome data, so it describes how quickly a team reached full lists and says nothing yet about results for patients.

The Royal College of Surgeons of England published updated guidance on 9 December 2025 asking for a Robotic Surgery Governance Group in every trust, minimum simulator training before patient cases, structured proctored cases with competency assessment, and formal supervision before independent practice. We first covered this guidance in our 22 to 28 September issue. Miss Nuha Yassin, the College's Council Lead for the Future of Surgery, Robotics and Digital Surgery, said: “The pace of innovation must be matched by rigorous standards.”

HTN reported on 2 October that Barts Health NHS Trust continues to roll out Oracle's Clinical AI Agent for ambient voice technology. More than 2,000 clinicians now have access, 1,350 have used it to generate patient letters from consultations, and 700 used it in September. About 8% of letters are still produced on the trust's legacy G2 speech system, and a limited number of G2 licences remain for selected users until the end of the financial year.

The trust's own account of its national trial involved about 250 colleagues. Over half saved at least five minutes per appointment, and two thirds reported better consultation quality. Letters to GPs are generated in real time, and the patient can see a copy before leaving the clinic. Neil Macdonald, the trust's Deputy Chief Medical Officer, said the trust listened to colleagues to ensure a smooth transition to the new approach to clinical documentation.

NHS England's guidance on ambient voice technology, published on 29 July 2026, sets five governance areas for boards: oversight and accountability, local governance, staff training, risk management and procurement. HTN's report notes that HSSIB has launched an investigation into the use of this technology in hospitals. The sources we reviewed give limited detail on how consent is recorded and how each letter is checked before it is sent.

HTN reported on 1 October that NHS England's board papers record concern about the absence of a single clear vision for the role of AI in the NHS, which risks delays and duplication. The board calls for proportionate and risk based arrangements at pace, to enable safe innovation and assurance across the lifecycle of AI systems. Its technology and innovation risk score has risen from 8 to 12. It names slow delivery across the Federated Data Platform, the NHS App, the single patient record and ambient voice technology, and cites limited provider capacity and internal financial, recruitment and approval processes.

The MHRA and Manchester University NHS Foundation Trust launched a Health Innovation Sandbox on 2 September to test AI enabled medical devices in real NHS settings, starting with tools that identify patients at higher risk of complications from chronic conditions. The launch included demonstrations of wearable products and ambient voice technology. A roundtable is planned for later in 2026, and the MHRA will invite expressions of interest from device developers for the next phase.

Health Innovation Minister James Frith said: “Innovation must never come at the expense of safety.” Mark Cubbon, Chief Executive of Manchester University NHS Foundation Trust, said patients should benefit from innovation as quickly as possible and should never have to choose between access and safety.