Add us as a preferred source
Joe McDonald, member of The Access Group health and care advisory board and former NHS consultant psychiatrist, argues that AI regulation is necessary – but only if regulators can keep pace with the technology.
There is often an argument made whenever a powerful new technology arrives that regulation and innovation pull in opposite directions. Artificial intelligence (AI) in healthcare is testing that argument harder than most at the moment.
I have spent the better part of 30 years watching the NHS wrestle with technology and innovation, and I have learned to be wary of both camps. The enthusiasts who think the rules are just an obstacle tend to leave a trail of expensive wreckage behind them. The purists who want everything nailed down before anyone touches it end up with nothing that works.
AI inherits this tension, with one difference that matters more than the rest. A large language model (LLM) can be revised in the time it takes an NHS board to agree its meeting agenda, and it may behave differently in one hospital compared to another fifty miles away.
The recent warnings from some experts that advanced AI could threaten humanity within a decade may yet prove overblown, but they capture something true about the speed. The technology is simply moving faster than those using it and those expected to oversee it.
Moon landings
If we consider one of the most significant innovation breakthroughs in our lifetime for a moment – putting a man on the moon. It wasn’t a triumph of careful regulation. NASA sat outside the framework that governed ordinary aviation and accepted risks that no safety regime would have waved through.
This milestone came precisely because the usual rules were not allowed to stand in its way. It is an uncomfortable thought for anyone drafting a rulebook, but caution and genuine progress have not always travelled together.
When it comes to the NHS, regulation is obviously necessary. The question has never been whether to regulate, but whether a regulator can move at the speed of the thing it is regulating.
And to its credit, the National Commission’s recent recommendations for AI demonstrate that it understands the problem it is dealing with.
You cannot regulate AI, and indeed many other digital solutions, the way you regulate a medicine or a conventional device. A drug is the same on the day it is licensed as it is a year later. An AI system can change after deployment, learn from new data, and usage can drift away from its originally approved purpose. Assessing it once, at the point of market entry, and issuing a certificate was never going to be adequate.
The report’s central recommendation is a shift towards proportionate, lifecycle regulation: continuous oversight across the life of a product rather than a single examination at the factory gate.
That is the right approach. The framework is also deliberately permissive. It accepts that these tools are already in clinical use and sets out to build sensible structures around them rather than to obstruct them. For a document of this kind, that is a mature and welcome position.
A good start, with real gaps
However, none of this means the work is finished. There are 44 recommendations, and a fair number ask the MHRA or the department to “consider” a course of action rather than commit to one. This is the beginning of a conversation rather than its end, and as beginnings go, it is a credible one.
What the report does not answer is how any of it will be funded, or who will actually carry it out. A regulator is almost by definition behind the curve, and asking an under-resourced one to keep pace with the best-funded technology in history is a tall order.
This is among the largest and least understood regulatory challenges the health service has faced, and it is being shaped largely by clinicians rather than by the people who build the models. Grounding the work in clinical reality is a genuine strength. It also places a premium on ensuring the authors are never quietly out of their depth on the technical detail. Likewise, emphasis on keeping the human in the loop is essential.
However, the Commission would do well to collaborate more closely with AI experts and vendors when it comes to implementation of the recommendations, if/when they are approved.
What it means on the ward
For frontline staff, the practical effect of the report is permission. Clinicians are already using ambient voice technology and AI assistance to draft correspondence and manage their workload. By acknowledging that, and taking a permissive line on adoption, the Commission gives them the confidence to carry on, with a framework forming around them rather than a cliff edge ahead. That matters more than the fate of any single recommendation.
The report will not clear away the Wild West on its own, and it cannot solve the problem of speed, because for now no one can. But it does describe the situation honestly and build around the work clinicians are already doing.
After two decades of difficult NHS technology projects, that is a more promising place to start than we are used to. Whether a regulator can keep pace with the technology it governs is the question the report leaves open, and it is the one that now matters most.
