A new stage for healthcare AI rules
The UK government has accepted all 44 recommendations made by the National Commission into the Regulation of AI in Healthcare, moving its blueprint from independent advice into a programme of government work. The decision commits the country to a proportionate, risk-based and lifecycle approach to AI-enabled healthcare products.
The announcement is paired with the opening of applications for the third phase of the Medicines and Healthcare products Regulatory Agency’s AI Airlock. The regulatory sandbox will work with eligible developers of AI-enabled medical devices to examine how products can be monitored safely after they enter use. Applications are open until November, and the first group of innovators is due to be selected that month.
The development is a policy commitment, not the arrival of a completely new regulatory code. The commission published its 44 recommendations on 10 September after gathering input from more than 12,000 patients, clinicians and members of the public. Initial priorities are set, and a fuller implementation roadmap, with responsibilities and timelines, is due by spring 2027.
Why monitoring cannot stop at approval
AI products used in healthcare may perform differently as data, clinical practice and the setting in which a tool is used change. A one-off assessment before deployment cannot, on its own, show how a device will continue to work in everyday care. The proposed lifecycle model therefore places greater weight on evidence collected while technologies are in use, alongside pre-market checks.
The regulator will develop a stronger approach to post-market assurance. That work is intended to improve the collection and use of real-world information about safety and performance, identify changes in quality or risk, and support a response when concerns emerge. The direction is to balance pre-market requirements with active oversight after deployment, rather than relying too heavily on a single assessment or passive incident reports.
The initial programme also covers how AI-enabled devices are classified and when a product qualifies as a medical device. The regulator plans to consult in 2027 on qualification and classification. It has also committed to issue draft guidance by December 2026 on managing changes to AI devices as they are updated or improved. These steps are still in development; the announcement does not itself alter the rules that manufacturers must currently follow.
Airlock phase will test oversight in practice
The AI Airlock’s third phase is focused specifically on post-market surveillance and lifecycle regulation. Developers will work with regulators and healthcare partners using real products and technologies to explore how monitoring can be carried out in practice. The work is expected to generate evidence and practical lessons for future guidance and policy.
Eligible developers can apply during the current window. A webinar for prospective applicants is scheduled for 22 October at 10am, with the first wave of participants expected to be chosen in November. The programme follows a pilot and a second phase, and has secured three years of additional government funding. It is a testing environment, not a general route that automatically authorises products for use in the health service.
The focus on ongoing oversight gives the sandbox a direct role in the wider reform effort. Testing how manufacturers and care organisations identify changes in performance could help clarify what information needs to be gathered, who should act on it and how problems should be escalated. Those questions matter because safe use depends on more than a product’s technical design: deployment conditions, clinical workflows and human oversight also shape how a tool functions.
Patients, staff and delivery milestones
Work beyond device regulation includes stronger patient involvement in decisions, improved routes to redress when standards of care fall short, and fairer access to benefits across different communities. The government will also explore staged authorisation pathways, under which promising tools might be used earlier with close supervision while further real-world evidence is gathered.
Responsibility is intended to be shared across manufacturers, healthcare providers, clinicians, regulators and other health-system partners. Nine priority areas for the coming year include workforce readiness and training, clearer conditions for safe deployment, and better reporting and learning from real-world use. These priorities will require coordination across the UK, including with devolved health authorities.
The next visible milestones are the December draft guidance on device changes, the 2027 consultation on qualification and classification, and the spring 2027 implementation roadmap. Until those measures are developed and adopted, the practical effect of the government’s decision is a defined direction and a set of work commitments. The third AI Airlock phase offers an early opportunity to test one of its central ideas: that assurance should continue throughout a technology’s working life.
