A framework built for a different era

Artificial intelligence is moving from pilot projects into everyday clinical decision making. Diagnostic support tools and treatment recommendation systems are now used across NHS and private settings. The legal framework that governs liability when these systems fail has not kept pace.

The Medical Protection Society, a UK based medical indemnity provider, has warned of increasing distance between the adoption of AI and the law. Its report, Closing the AI Liability Gap, argues that clinicians are currently exposed to disproportionate legal risk when AI systems cause harm.

Why clinicians carry the risk

The problem lies in how AI is classified. The Consumer Protection Act 1987 was designed for physical products. It was never written with software or algorithms in mind. As a result, AI systems often fall outside its scope. Developers, manufacturers and suppliers of AI tools are effectively shielded from product liability rules that would normally apply.²

This creates a default position. If a patient is harmed by a defective AI system, the practical route to compensation becomes a clinical negligence claim against the clinician who used it. The NHS and privately practising doctors then absorb costs that arguably belong elsewhere in the supply chain.

This is not solely a UK problem. Similar gaps exist in the EU’s proposed Product Liability and AI Liability Directives, particularly for opaque systems whose reasoning cannot easily be tested. Patients may struggle to prove a defect exists at all, let alone identify which party caused it.

The case for reform

The Medical Protection Society has called on the UK Government to legislate so AI systems are explicitly classified as products. This would bring AI within scope of established product liability principles, in the way pharmaceuticals and medical devices already are. Responsibility for harm could then be shared fairly across developers, manufacturers, importers and clinicians, rather than defaulting almost entirely to the end user.

Sarah Townley, Deputy Medical Director at The Medical Protection Society, has described the current framework as inequitable and unfairly exposing clinicians. She argues that clear, shared liability would also build trust in AI and encourage developers to prioritise safety at the design stage, rather than leaving end users to manage risk after deployment.

The Law Commission is currently examining the liability of digital products and new technologies. A public consultation is expected later in 2026. This represents a genuine opportunity to close the gap before AI becomes further embedded in clinical pathways.

What this means in practice

Until legislation changes, clinicians should treat AI recommendations with the same scrutiny applied to any other clinical tool. Documentation should record how an AI output was used and why it was accepted or rejected. Reliance on an algorithm does not remove professional judgement or accountability under current negligence law.

The direction of policy travel is clear. Liability should follow control and knowledge, not simply proximity to the patient. Until reform arrives, clinicians remain the most exposed party in the chain. Understanding this gap, and documenting AI use carefully, is the best protection available in the meantime.