Artificial intelligence is increasingly entering clinical pathways. It supports diagnosis, helps monitor patients, assists decision-making, organises complex clinical data, and promises a more personalised form of medicine. This development is particularly relevant in areas such as dementia, frailty, and ageing, where early detection, continuous monitoring, and decision support can significantly improve care.
At the same time, these very domains make it clear that the risks of AI in healthcare are not merely technical. They do not concern only model accuracy, data quality, or cybersecurity. They also involve trust, autonomy, the clinician–patient relationship, the role of caregivers, the emotional burden on families, and the capacity of vulnerable individuals to understand, accept, or contest AI-supported decisions.
The ethical literature on AI in healthcare has already identified a wide range of risks: bias, opacity, lack of explainability, unclear accountability, data misuse, over-reliance on automated systems, and erosion of professional autonomy. However, more recent scholarship highlights an important limitation: general principles, while necessary, are not sufficient. What is needed is a deeper understanding of how these risks materialise in real-world care settings, within clinical workflows, and across the lived experiences of patients and professionals. In this regard, Ratti and colleagues argue for a “situated ethics” of AI in healthcare – one that focuses not only on systems, but on the contexts in which they are actually used.
This shift is particularly critical in dementia and frailty care. In these settings, individuals may experience fluctuating decision-making capacity, cognitive impairment, complex care needs, and a strong dependence on family members or professional carers. A risk score, an automated alert, or an AI-generated recommendation is never just an output: it can reshape how a person understands their condition, how families negotiate responsibility, and how clinicians communicate uncertainty and prognosis.
It is within this context that the work carried out in the COMFORTage project becomes especially relevant. COMFORTage is a Horizon Europe initiative aimed at improving the diagnosis, monitoring, and prevention of dementia and frailty through an integrated, multidisciplinary approach. By combining clinical expertise, technological development, and social sciences and humanities, the project seeks to enhance both the effectiveness and the acceptability of AI-enabled healthcare solutions.
Within COMFORTage, CyberEthics Lab applied the ETHAI methodology to analyse AI-related risks not as abstract categories, but as situated, relational, and evolving phenomena. The approach combined three complementary layers: analysis of the scientific literature, clinician workshops, and the co-creation of narrative scenarios.
The first layer focused on identifying domain-specific risks through a targeted review of literature on AI in neurology, ageing, and dementia. Beyond the well-known categories of AI ethics, this body of work highlights additional concerns: mental privacy and cognitive liberty, misinterpretation of behavioural data, medicalisation of ageing, overdiagnosis, digital exclusion, challenges in informed consent, and the potential erosion of relational care. In these contexts, systems that monitor behaviour, speech, sleep, or social interaction can generate clinically useful insights – but also risk turning ordinary variations in daily life into signals of pathology.
The second layer, based on clinician workshops, translated these abstract concerns into concrete practice-based issues. Healthcare professionals emphasised risks such as false precision (outputs that appear authoritative but mask uncertainty), difficulties in communicating probabilistic results to patients, increased workload, alert fatigue, unclear responsibility, and tensions between clinical judgement and algorithmic recommendations. These insights highlight a key point: ethical robustness depends not only on system design, but also on how AI integrates into workflows, decision-making processes, and patient interactions.
The third layer introduced a more innovative element: narrative scenario-building. Clinical partners were invited to construct short, plausible stories describing how AI systems developed within the project might operate in everyday care settings. This was not an exercise in fiction for its own sake, but a structured method for anticipatory ethical reflection.
This approach draws on the tradition of narrative medicine, which emphasises the role of storytelling in understanding clinical complexity, uncertainty, and moral decision-making. Research has shown that narrative approaches can enhance clinicians’ ability to interpret situations, engage with patient perspectives, and navigate ethical tensions. In the context of AI, narrative methods offer a way to explore not only what systems do, but how their effects unfold over time.
In COMFORTage, narratives revealed aspects of risk that remained largely invisible in both literature and workshops. Risks were no longer perceived as static properties of AI systems, but as processes evolving within relationships, routines, and organisational dynamics. For instance, repeated alerts could gradually increase anxiety within families; risk scores could influence a person’s sense of identity; AI recommendations could generate conflicts between professionals or create uncertainty about responsibility; and digital tools could unintentionally replace meaningful human interaction.
What emerges from these narratives is a deeper understanding of the temporal and relational nature of AI risk. A minor technical issue, such as an ambiguous alert, can have disproportionate psychosocial consequences in a fragile care context. A probabilistic prediction can become a source of distress rather than guidance. A tool designed to support clinicians may, if poorly integrated, increase organisational burden rather than reduce it.
Narrative methods therefore provide something that traditional ethical frameworks struggle to capture: the trajectory of risk. They show how risks accumulate, interact, and transform over time, and how they become embedded in the moral and emotional fabric of care.
The work conducted in COMFORTage reinforces a key insight: the ethics of AI cannot be reduced to compliance with general principles or regulatory checklists. It must address how technologies reshape care practices, relationships, responsibilities, and lived experiences – especially in contexts of vulnerability.
Importantly, the narrative and co-creative approach does not replace technical or regulatory analysis. Rather, it complements it. It enables clinicians to act as active interpreters of AI’s implications, brings to light risks that are difficult to anticipate analytically, and supports the translation of ethical concerns into more concrete design and governance requirements.
In this sense, COMFORTage illustrates how ethics can be embedded within innovation itself – not as an external constraint, but as an integral part of design and implementation. Developing trustworthy AI in dementia and frailty care requires more than ensuring that systems work correctly. It requires understanding what changes, for whom, and under what conditions.
Ultimately, if AI is to support care in meaningful and responsible ways, we need methods that can engage with the complexity of real-world contexts. Narratives – grounded in clinical experience and co-created with practitioners – offer a powerful way to anticipate these complexities and to design technologies that are not only effective, but also aligned with the values and needs of those they are meant to serve.
References
- Ratti, E., Morrison, M. & Jakab, I. Ethical and social considerations of applying artificial intelligence in healthcare—a two-pronged scoping review. BMC Med Ethics 26, 68 (2025).
- Volpini, L., Prosseda, V., Morpurgo, F., Glišovic Krivec, Š., Krivec, D. (2025). The ETHAI Methodology. In: Papaleonidas, A., Pimenidis, E., Papadopoulos, H., Chochliouros, I. (eds) Artificial Intelligence Applications and Innovations. AIAI 2025 IFIP WG 12.5 International Workshops. AIAI 2025. IFIP Advances in Information and Communication Technology, vol 754. Springer, Cham.
- Ghenimi, N., Govender, R. & Moodley, K. The paradox of artificial intelligence (AI) and narrative-based medicine: challenges and potential for enhanced patient care. AI & Soc 41, 251–257 (2026).
