AI tools are already embedded in NHS clinical workflows. One AI system for chest X-rays is now running across more than 40 NHS Trusts. Another is being used to spot breast cancer in mammograms, with one trial reporting it found around 12% more cancers than standard double-reading. The government's own 10-year health plan sets a target of making the NHS the most AI-enabled health system in the world. That is the direction of travel, and it is fast.
The legal structure underneath all of this has not moved. Under current UK law, if an AI tool makes a wrong call and a patient suffers, the doctor who used the tool or the hospital that deployed it carries the legal responsibility. The software maker is largely protected by product liability rules that were written for physical goods, not for software that changes its behaviour over time. Proving that an AI system was "defective" in the legal sense is extremely difficult, because these tools constantly update and adapt.
The Medical Protection Society, one of the main bodies that insures and defends doctors in the UK, is now warning ministers directly. Their concern is that clinicians are becoming what they describe as a "liability sink": the single point where all legal risk lands, regardless of who actually made the error. The UK's medical regulator, the GMC, reinforces this by expecting doctors to critically assess any technology they use. If a doctor follows an AI recommendation that turns out to be wrong and did not document their own independent reasoning, they could be found negligent.
This creates a genuinely difficult position for any clinician. The government is pushing AI adoption. Hospital managers are deploying these tools. But the individual doctor standing in front of the patient is the one who will face court if something goes wrong, even if the AI gave the instruction.
Ninety percent of doctors who currently avoid AI cite this legal exposure as their top reason. That is not technophobia. That is a rational response to a system where the risk and the reward are badly misaligned.
The problem is not unique to the UK. Globally, as of mid-2025, no AI system had been held legally responsible for patient harm in any court anywhere. The law everywhere treats AI as a tool, not as an actor. In the US, courts are working case by case, and some states are starting to draft specific AI liability laws. The EU has gone furthest, introducing both an AI Act that labels medical AI as high-risk and a liability directive designed to make it easier for patients to seek compensation from AI developers. The UK, outside the EU since Brexit, is building its own framework: a new regulatory structure from the MHRA is expected in 2026.
For anyone running a business that deals with healthcare, insurance, or professional services, the direction of this debate matters. The core question being settled right now is: when an automated system causes harm, who pays? The answer affects insurers who underwrite medical professionals, procurement teams buying health technology, and any organisation deploying AI tools to support decisions that affect people.
The most likely outcome is a shared liability model, where responsibility sits across the clinician, the hospital, and the software developer, weighted by how much control each had over the decision. That is roughly where self-driving car law landed. But that framework does not exist yet in healthcare, and until it does, the doctor in the room carries the risk alone.