Search the hub

This World Health Organization (WHO) report analyses the landscape of digital health competency frameworks and standards developed to strengthen health workforce education and practice in increasingly digitalised health systems. It examines how digital health competencies are defined, structured and applied across different professional, educational and organizational contexts, with a focus on supporting the effective use of digital technologies in health service delivery and learning. The report is situated within the broader context of accelerating digital transformation in health, including increased demand for telemedicine, remote learning and digital tools following the COVID-19 pandemic and other global disruptions. Based on a narrative review of the literature, the report identifies and maps competency domains across existing frameworks, highlighting common competency clusters related to patient care, data, informatics, communication, technical proficiency, digital professionalism and administration. It discusses methodological approaches used in framework development, key implementation challenges and factors influencing adaptation and sustainability. The publication also outlines policy considerations to support the development, contextualization and use of digital health competency standards by policy-makers, educators and health institutions to strengthen digital competencies within the health workforce.

AI mental health self-help tools are growing fast but protection for the people using them isn't keeping pace. This paper from David Gilbert and the Centre for Mental Health finds people’s use of AI to support mental health has outpaced the development of robust mechanisms to mitigate problems. Oversight is uncoordinated, and there are significant gaps in evidence, accountability and patient safety. While these tools may improve access and affordability for some, the paper warns that the benefits won't be distributed evenly - and that the risks of generative AI mental health systems are likely to fall disproportionately on people who are already vulnerable. Large language models can also absorb and repeat patterns of structural discrimination, reinforcing stereotypes or invalidating certain identities.

Physical AI refers to artificial intelligence (AI) systems that operate in and interact with the physical world, rather than existing only in software or digital environments. Physical AI typically involves the combination of AI models with sensors, actuators and other control systems that allow models to act upon real-world environments, taking models from the realm of bits to the realm of atoms. With AI, advanced physical systems can now perceive the environment, reason with the power of a large language model (LLM), act accordingly, and then learn from the outcome of that action. This IBM article explains more.

These two reports summarise findings from the National Commission into the Regulation of Artificial Intelligence (AI) in Healthcare’s research and engagement activities and call for evidence. The Commission’s purpose is to advise the Medicines and Healthcare products Regulatory Agency (MHRA) on improving its regulatory framework and to accelerate safe access to AI in healthcare and across the NHS. You can read a summary of Patient Safety Learning’s response to this call for evidence here. The work brought together evidence from patients and the public, healthcare professionals, industry, academics and wider health system stakeholders through public polling, surveys, stakeholder engagement, deliberative research, an open call for evidence, a public Ask Me Anything session and insights from the MHRA’s AI Airlock programme. Thorough analysis of this evidence, 10 key findings have been identified. The report summarises these as follows: 1. There is a clear call for a proportionate, lifecycle-based approach to regulation Stakeholders noted that the current framework, which is designed for more static medical devices, is not well suited to iterative and adaptive AI systems. Across groups, stakeholders called for a proportionate approach that is risk-based, considers patients’ safety and fairness, with clear practical guidance and addresses existing duplication and fragmented oversight. Stakeholders also underlined the importance of strengthening clinical evidence requirements, with strong support for enhancing post-market surveillance and improving coordination. With a more proportionate approach seen as essential for balancing innovation with patient safety. 2. There is strong consensus for significant regulatory reform Across respondent groups of healthcare professionals, healthcare providers and industry, most people said that the existing regulatory framework needed “significant reform” but did not need a “complete overhaul”. Amongst patients and the public, the number of respondents calling for “significant reform” and a “complete overhaul” were similar, with 34% asking for “significant reform”, and 35% for a complete overhaul. 3. There was broad consensus that AI systems will increasingly require continuous post-market surveillance and monitoring Several stakeholders highlighted the need to upgrade current approaches to post market surveillance and monitoring, so they are better suited to AI systems. There was strong consensus that performance and risk cannot be adequately assessed through one-off approvals alone but instead require ongoing, real-world oversight across the lifecycle. Through qualitative evidence, stakeholders called for a more continuous and ongoing approach which helps track performance, monitor safety, and manage compliance across the AI system lifecycle. They also suggested that upgraded approaches need to help manage performance drift, validate performance in real world settings, and track changes in performance over time. 4. Responsibility should be shared across the system, with each individual and institution understanding their essential role and responsibilities There was strong consensus that accountability should not rest with a single person or institution, with respondents favouring a model which better distributes liability across the lifecycle. Patients and members of the public called for a comprehensive approach to accountability that addresses current gaps, healthcare professionals stressed that clinical accountability should be maintained whilst healthcare providers emphasised the need for robust governance structures and clear organisational responsibility. Stakeholders also highlighted uncertainty in how roles, responsibilities, and liability are defined and applied in practice. There were differing views on where liability should sit when an AI system causes or contributes to harm. Some respondents believed that liability should sit with the healthcare professional using the AI system. Another group of respondents argued that liability should sit with the healthcare provider who deploys the AI system. Others suggested that liability should sit with manufacturers, given their role in developing the technology and then maintaining their AI system’s performance. Across responses, there was a consistent emphasis on the need for greater clarity and consistency in how liability is allocated. Many respondents called for structured approaches to distributing liability that reflect the roles of different actors, including manufacturers, healthcare providers, and healthcare professionals. Suggested approaches included shared or distributed liability models that apportion responsibility based on specific circumstances. Stakeholders noted that clearer and more consistent frameworks would help address uncertainty and support the safe use of AI systems in healthcare. 5. Human oversight and responsibility for clinical judgment should be retained There was strong consensus from respondents that AI systems should continue to augment the work of professionals and should not be fully responsible for clinical decision making. Patients and the public emphasised the importance of human involvement in their care, including expectations that clinical decisions involving AI should be checked and validated by a human clinician. Healthcare professionals and professional bodies highlighted the risk of over-reliance on AI outputs at the expense of professional judgement. Industry respondents were supportive of ‘human-in-the-loop' safeguards. 6. Transparency and explainability will be key for the ongoing deployment of AI systems The ability to easily understand how an AI system works and to interpret its outputs will be key for building trust, enabling deployment, and ensuring the safety of an AI system. Patients, public and professionals advised that explanations of AI system outputs need to be clear, and providers called for greater transparency in the procurement process for sourcing AI systems. Industry organisations commented on the need for clearer and more structured regulatory documentation. 7. Data access and use is central to the role of AI in healthcare moving forward Respondents to the Call for Evidence noted that healthcare data is simultaneously an enabler and a barrier to the development and deployment of AI systems in healthcare. Patients and public expressed strong concerns about current approaches to consent for data access and how data is used by commercial entities. Some respondents cited governance and compliance burdens and fragmented data infrastructure as key barriers to development and deployment. Industry respondents called for clear and robust frameworks for accessing data including shared data governance templates and clearer guidance on data standards. 8. There is a need for robust training and improved AI literacy The Call for Evidence found a clear view that robust, ongoing training and clear understanding of AI in healthcare is critical for safe adoption. Healthcare professionals highlighted the risks of a lack of AI-specific training can bring such as increased risk of automation bias. Healthcare providers called for more structured workforce training on AI moving forward. Industry respondents advised that training is also needed for individuals who oversee the governance of AI systems in healthcare. 9. There is a need to improve incident reporting and learning mechanisms There were widespread calls for standardised reporting mechanisms for AI systems. Patients and public called for greater transparency and accountability over where AI is involved in care, including clearer communication when things go wrong. Healthcare professionals raised concerns about underreporting of safety incidents in healthcare more broadly, noting that workload pressures are a significant contributing factor. Responses also suggested limited awareness amongst some healthcare professionals that the existing Yellow Card scheme already applies to medical devices, including AI enabled devices. Healthcare providers highlighted the operational challenges of implementing incident reporting consistently across different settings. Industry respondents called for clearer guidance on how incident reporting should work within AI specific post-market surveillance frameworks. Several respondents also proposed improvements to surveillance and monitoring approaches, including establishing a national reporting system for AI incidents and providing guidance for healthcare professionals on what to report. 10. Patient and public engagement, trust, and communication will continue to be key for the deployment of AI systems. Through the Call for Evidence, trust emerged as a core enabler of AI adoption in healthcare. Patients and the public called for consistent involvement, consent, and clarity over the role of AI systems, whilst professionals highlighted the need to take a proportionate approach to explaining how AI is being used to patients. Providers advised that clear and consistent transparency and communication frameworks are needed whilst industry respondents recognised that trust is key for the uptake of AI systems in healthcare.

Medical device makers have been rushing to add AI to their products. While proponents say the new technology will revolutionize medicine, regulators are receiving a rising number of claims of patient injuries. This Reuters Special Report investigates some of the hazards associated with AI-enabled medical devices, including errors in a navigation system integrated into a medical device used in ENT surgery, AI software used for prenatal ultrasound scans that misidentified fetal body parts and AI assisted heart monitors that failed to recognise abnormal rhythms.  Issues with the capacity of the U.S. Food and Drug Administration (FDA) to review the flood of new AI-enabled medical devices are also raised, as well as concerns that the FDA's traditional approach to regulating medical devices may no longer be fit for purpose.

The law has always struggled to keep up with technological change. With AI, the pace of change is so rapid that this gap feels less like a step and more like a widening gulf.   A recent White Paper, produced through a collaboration between the MPS Foundation, York University’s Centre for Assuring Autonomy, and the Improvement Academy at the Bradford Institute for Health Research, highlights how clinicians could find themselves exposed when their decisions are influenced by AI recommender systems. Such systems analyse patient data and suggest personalised treatment plans, diagnoses or medications.  There are also concerns about who might be held liable in the event of a claim relating to AI scribes, automated documentation assistants, triage algorithms, and other forms of clinical decision support. These all share a common feature: they shape clinical reasoning, records, and workflows without taking autonomous responsibility for the outcomes. Under the current legislative framework, there is a risk that doctors could be held wholly liable if an AI suggestion turns out to be wrong and they have followed it. That’s because the existing product liability regime was never designed with AI in mind.   This paper from Medical Protection aims to set out the challenges we expect clinicians will face, and the action policymakers can take now to make sure AI delivers benefits without leaving doctors unfairly exposed. 

This guide highlights key considerations for audit and risk assurance committees when overseeing the planning, deployment and scaling of artificial intelligence (AI) within public sector organisations. It draws on National Audit Office (NAO) findings, the UK Government’s AI Playbook, and lessons from digital transformation programmes across government. This guidance includes: where AI is used in government areas that organisations need to consider areas of focus and suggested questions to ask.