UK to Replace One-Off Approval of Medical AI with Lifetime Monitoring After Ministers Accept All 44 Commission Recommendations
The UK government has accepted all 44 recommendations of the National Commission into the Regulation of AI in Healthcare. The central change is a move away from one-time approval of AI-enabled medical devices and towards monitoring them throughout their working life.
Under the current model, a device is assessed before it reaches the market and is then largely left to operate. That approach sits uneasily with software that learns and changes. A machine learning tool can alter its behaviour after updates or as it processes new clinical data, so a pre-market check may say little about how it performs months later. The government's response commits regulators to assessing real-world performance over time.
The Medicines and Healthcare products Regulatory Agency (MHRA) will carry most of the work. It will publish draft guidance by December 2026 on how updates to evolving AI devices should be managed. That guidance is expected to address how far a manufacturer may change a model before a fresh assessment is needed. A public consultation on the initial classification of AI-enabled devices across clinical settings will follow early next year.
Testing of these ideas will take place through the MHRA's AI Airlock, a regulatory sandbox in which developers, regulators and health system operators trial products under supervision. The agency has opened applications for the third phase, which has been funded by the government for three years. This phase will examine post-market surveillance and continuous monitoring of AI tools already in use. A webinar for prospective developers is scheduled for 22 October 2026, and the first cohort will be chosen in November.
The response also points to a change in how promising tools reach patients. Ministers are exploring staged authorisation pathways, which would allow AI products with high potential to enter clinical use earlier, provided they remain under close observation. Supporters of such a model argue that it could shorten the wait for useful technology. Critics of staged entry tend to raise the risk that a tool is exposed to patients before its limitations are fully understood, which makes the strength of the monitoring system central to the design.
Patient-facing commitments sit alongside the regulatory changes. The government has said it will increase public transparency about when AI is used in care and improve the routes by which patients can seek redress if something goes wrong. It has also committed to ensuring that the benefits of the technology reach underserved groups. Concerns about bias in AI systems have been raised repeatedly in health settings, since a tool trained on data from one population can perform less well on another. Continuous monitoring could in principle reveal such gaps after deployment, though the response does not yet set out how performance across different groups will be measured.
Delivery will be overseen by a new cross-agency Programme Board. Its work is divided across ten priority areas, and it is responsible for producing a full implementation roadmap by spring 2027. That document is expected to set out which agency is responsible for each action and by when. Until it appears, many practical questions remain open, among them how the monitoring will be funded, what data manufacturers and NHS organisations will be required to share, and how quickly regulators can act if a device begins to underperform.
The pace of the programme is deliberately staged. Guidance, consultation, sandbox trials and the final roadmap are spread across roughly the next eight months, which gives the MHRA time to test its approach before it becomes binding. Developers will watch the December guidance closely, since it will provide the first detailed indication of what ongoing obligations look like in practice.
For now, the commitment is to the principle. Accepting every recommendation signals that ministers regard the existing framework as unsuited to adaptive software. Whether the new system delivers faster access and stronger safety in equal measure will depend on details that have yet to be published.
