What the research found
The Commission, set up by the MHRA in September 2025 and chaired by practising NHS doctors, gathered evidence from patients, clinicians, healthcare leaders and developers over a year before publishing its recommendations. It proposes that new AI models be given staged authorisations — deployed first under close supervision and tight guardrails, like L-plates for learner drivers, before earning fuller approval; that AI-enabled devices be monitored continuously in the real world rather than cleared once at a point in time; that the public be able to search safety information and adverse incidents for a specific device; and that the MHRA get stronger enforcement powers. Companion deliberative research by the Health Foundation, published the same day, found public support for AI in care is real but conditional, with accuracy the top priority and human oversight treated as a precondition rather than a nice-to-have.
A national regulator is now formally proposing what providers have had to improvise: that an AI tool earns trust in stages and stays monitored for its whole working life, rather than being declared safe once and left alone.