Clinicians are being lined up as a "liability sink" for AI errors, the Medical Protection Society warns, as it presses ministers to reroute negligence claims to the developers.
When an AI tool reads a chest X-ray and misses a tumour, the doctor who signed off on the report, not the company that built the algorithm, is the one a patient can sue for clinical negligence. That is the structural problem the Medical Protection Society, the body that defends doctors accused of wrongdoing in the UK, is now asking ministers to fix before AI use in the NHS outruns the law that governs it (the Guardian).
MPS has published a report warning that doctors and the NHS itself could face negligence claims for harm that originates in an algorithm they did not design, cannot fully inspect, and may not be allowed to override. The society's framing is pointed: under current law, the clinician is the default defendant, and MPS calls that role a "liability sink" for AI-made mistakes (the Guardian).
The pressure point is that this is no longer a hypothetical. AI tools are being deployed inside NHS clinical pathways — a development the MPS report documents — and once those tools sit inside the diagnostic pathway, an error in a model's output becomes, in legal terms, an error in the clinician's decision to rely on it. That is the gap between product behaviour and professional responsibility, and it is what the present legal framework does not separate.
MPS's policy answer is to redraw that line. It wants AI tools used in medicine reclassified as products, so that when one fails, the claim travels under product liability law, against the developer, rather than under clinical negligence law, against the doctor. The shift changes three things at once: who is named in the claim, what the claimant has to prove, and who ultimately pays.
None of this is an argument against putting AI into NHS workflows. The tools are being deployed because they can speed up triage and reduce administrative load, and clinicians want them to work. The point is that adoption is outpacing the legal rails that would make adoption safe, and a clinician facing a negligence claim several years after the fact is not a sensible backstop for a vendor's quality assurance.
What is still missing is a government response. MPS speaks for the workforce side of the risk, but patients retain a non-negotiable right to recourse when something goes wrong. The design question for ministers is therefore not whether to extend liability, but to whom: how fault is assigned when an AI system is part of a clinical pathway, what audit trails clinicians and vendors will be required to keep, and what indemnity arrangements follow from reclassification. Until those answers are on the page, the law will keep routing the consequences of an algorithmic mistake to the person with the least control over the algorithm.