AI in the NHS

Approximate read time: 5 minutes

AI has a range of potential applications in healthcare, from administrative tasks, such as organising appointments, drafting letters and transcribing patient notes, to clinical applications, including triage, diagnosis and planning treatments.^[1] These systems can fully automate processes or be used as tools under human supervision with the aim of reducing costs and improving health outcomes. Some leaders in healthcare have argued that AI will be essential for the NHS to meet rising demand, although others have stressed a need for realism, pointing to current practical, regulatory and ethical challenges that may limit its use.^[2]

The government’s 10 year health plan for England seeks to “make the NHS the most AI-enabled health system in the world”.^[3] AI is already being used in some capacities in the NHS.^[4] For example, a recent Microsoft Copilot trial involving over 30,000 NHS staff reportedly saved each staff member 43 minutes a day on average through automating administrative tasks.^[5] Current clinical applications include identifying disorders from blood samples, and detecting early signs of lung cancer in X-ray and CT scan images.^[6] Work on a new NHS-wide cloud computer system is ongoing to further enable the use of AI-screening technologies across the country.^[7]

1. Risks and public sentiment

Although the government has ambitious plans for AI in the NHS, public opinion is divided. A 2024 survey from the Health Foundation found that 61% of people were supportive and 28% were not supportive of the use of AI in the NHS for administrative purposes, dropping to 54% supportive and 33% not supportive for clinical applications.^[8] Some of the key concerns raised in this and other surveys are explored below.

1.1 Accuracy and transparency of decision making

The most commonly reported concerns in public surveys from the Health Foundation and Royal College of Radiologists were related to the accuracy of AI decision making.^[9] AI systems can make mistakes which present safety risks in healthcare settings, such as missing diagnoses.^[10] Studies also suggest that people are less accepting of errors made by AI than by humans.^[11]

In some cases, AI can make systematic mistakes that exacerbate existing inequalities. For example, AI performed worse at identifying skin cancer in people with darker skin due to poor representation in training datasets.^[12] The NHS’s AI ethics initiative, which concluded in 2025, funded research into how AI could be designed and deployed to counter such inequalities, publishing guidance and recommendations for its ethical use.^[13]

There is also the question of who should be liable for AI’s mistakes: should it be the healthcare professionals applying the tool or the AI developers?^[14] This is further complicated by the fact that the decision making processes of AI systems are not transparent, making it difficult to identify the cause of errors.^[15] The Law Commission recently explored the possibility of granting AI systems legal personalities to solve the problem of liability, speculating that this “potentially radical” option may become increasingly salient as AI systems advance.^[16]

1.2 Impacts on the human dimension of care

The Health Foundation survey found that the most common concern among NHS staff, and third most common for the public, was that AI would make healthcare more impersonal.^[17] This could be through decisions made by unempathetic AI tools or patient-facing systems replacing human interactions. The latter could include chatbots or generative AI voice agents having conversations with patients to triage symptoms, perform daily check-ins or track medication adherence.^[18]

Others, including the government, argue that the use of AI for administrative tasks could be an opportunity to make healthcare more personal.^[19] A recent study found that resident doctors spend four hours on admin for every hour with patients.^[20] Automation of administrative tasks could free up time to spend with patients.^[21]

1.3 Data privacy

The training and application of AI require the centralised collection and storage of large amounts of highly sensitive health data. Data used to train AI systems can be anonymised, yet there is a risk of re-identification if not done correctly.^[22] A 2024 NHS England survey found that people trusted “technology companies developing apps, services or AI” the least out of nine organisation types, with only 27% saying they trust these companies with patient data.^[23] The most common concern was cyber-attacks, such as the 2024 Synnovis data breach.^[24] A key challenge for regulators will be to balance the need for data protection and data access for developing and testing AI algorithms.^[25]

2. Developing a new regulatory framework

Some senior figures in healthcare have emphasised a need to build trust in AI through public engagement and effective regulation.^[26] The Medicines and Healthcare products Regulatory Agency (MHRA) has established the national commission into the regulation of AI in healthcare to review current regulations and provide recommendations for a new regulatory framework for AI in healthcare.^[27] One member of this commission, Dr Jennifer Dixon, chief executive of the Health Foundation, has argued for the need for a more flexible approach to regulation, moving from a one-off pass/fail process to implementing systems with ongoing reviews.^[28] The commission’s recommendations are expected to be published in 2026.^[29]

---

Image from Freepik.