Expert comment: AI is helping the NHS, but who is responsible when it gets things wrong?

Craig comments: “Artificial intelligence is already becoming part of everyday NHS care, from analysing medical images to producing consultation notes. The National Commission into the Regulation of AI in Healthcare has made 44 recommendations for a regulatory system that is “safe, fast and trusted.”

“The report recognises a fundamental problem: unlike many conventional medical devices, AI can be updated, adapt to new information and perform differently between hospitals or patient groups. It therefore argues that one-off pre-market assessment is insufficient. AI-enabled medical devices should be monitored throughout their working lives, with appropriate version control and traceability where they influence patient care.

“This is crucial when something goes wrong. The report identifies concerns that the existing negligence framework can turn doctors and healthcare providers into perceived “liability sinks,” with responsibility transferred to them without sufficiently recognising the influence of AI systems and wider system design. We argue that human oversight is meaningful only if clinicians understand the technology, receive appropriate training and have sufficient time and information to question its output.

“The issue is pressing. BBC analysis found that A&E-related clinical-negligence claims in England rose by 41% over five years. While it has been reported that compensation payments for A&E mistakes increased from £231 million in 2015 to £475 million in 2025. AI may help identify deterioration or interpret scans, but it cannot compensate for inadequate staffing, beds or time. Introduced without appropriate safeguards, it could compound existing risks.

“The Commission recommends continuous monitoring, clearer allocation of responsibility, improved patient information and staged authorisation for appropriate AI-enabled medical devices. It also calls for a proportionate, system-level approach that addresses patients’ reasonable expectation of being informed and their ability to opt out where possible or appropriate. Questions remain about consent, liability and access to redress.

“These recommendations do not themselves change the law. A formal government response will follow. Their value will ultimately depend on whether broad principles become clear and effective duties, and whether regulators and healthcare organisations receive the expertise and resources needed to make them work.”