Autonomous driving in the United States, from personal cars to public taxis, has been ahead of the curve over its period of emergence. This outcome is aided by the preferable environment that typical American city roads offer: geometric grid planning combined with vast, flat terrain.

However, tackling US tracks is far different to dealing with the ambiguous streets of London. The density and proximity of vulnerable road users, and external factors like weather, makes safe driving in the United Kingdom reliant on social interaction, such as eye contact, hand gestures or flashing lights.

Health and care is just like a London street: messy, adaptive and full of “corner cases” that require more than just rigid rules. Navigating health and care passages demands common sense, human oversight and enormous testing and regulation before new tools can be implemented.

Within their recent publication in Nature Medicine: Building a code of conduct for AI-driven clinical consultations, a team of experts form interdisciplinary research.

To gather further insight into this formative project, CERSI-AI interviewed a group of the authors to discuss the paper.

During our interview, Dr Ernest Lim observes that AI is “doing amazing things” following up by posing the question “Why is that not reality, right? Why are patients not benefiting from that yet?”

The group discuss this disjunction. Research revealed that current AI evaluations prioritise quantifiable diagnostic accuracy, rather than relational skills essential for patient-centred care.

While intelligent adaptation is typical for advanced AI tools, safety challenges arise from potential “corner cases,” defined by rare occurrences that human clinicians can resolve through emotional intuition. Engineers encountered a similar problem when automating driving. Safe driving comes from more than just correct navigation, but intrinsically human connections like communicating, empathising and planning.

AI’s inability to harness genuine patient rapport causes problems, Helen Monkhouse concurring “how [do] you know that it will deal with every situation?”

Dr. Ernest Lim summarises the importance of their work: “Everyone’s rushing to replace clinicians, but no one’s really stopping and saying, ‘What are we actually replacing?'”

Adopting principles from the vehicle industry, such as the shared language of the highway code, and self-driving car evolution and development, the team constructed a two-layer framework for clinical consultations. This system can be applied to regulate AI use and discern behavioural necessities for quality patient experiences.

Layer 1: “Cure”: The instrumental layer. Technical tasks like eliciting symptoms and creating treatment plans.

Layer 2: “Care”: The relational layer. The “softer things” like recognising distress, offering empathy, and building trust.

The paper recognises Michael Balint’s (1955) ‘Doctor as the Drug’ as a foundational concept within the framework, acknowledging that therapeutic sides in clinical consultations is as powerful as any AI tool.

• Lim et al (2026) Building a code of conduct for AI-driven clinical consultations Nature Medicine.

Regulation is certainly the next step towards AI medical devices actually benefiting patients’ health and care safely.

It is essential that any automation in health and care follows the incremental, risk-based road that self-driving car development is paving. As such, virtual testing can be used to ensure that real-world deployment of these technologies arrives safe, swift and soon.

In terms of the trajectory of the health system, the authors suggest that viewing AI as an “agent” alongside humans faces towards a synchronised, collaborative future.

Despite the focus on transferring cross-industry learning to health and care, Ibrahim Habli draws on his experience to recommend that “perhaps other industries can learn a lot from healthcare, especially when it comes to resilience.”

Finalising our interview, Ibrahim Habli articulates a core question that organisations like CERSI-AI will strive to address: “How do you regulate a team of human and artificial agents coming together?”

Read the full paper in Nature Medicine to explore the behavioural criteria for safe AI in clinical consultations:

• Lim et al (2026) Building a code of conduct for AI-driven clinical consultations Nature Medicine.

Follow us at CERSI-AI as we work to bridge the gap between extraordinary technologies and patient benefit.

Connect on:

Balint, M. (1955). The Doctor, his Patient, and the Illness. The Lancet, 265(6866), pp.683–688. doi:https://doi.org/10.1016/s0140-6736(55)91061-8.

Lim, E., Thirunavukarasu, A., He, Y.V., Monkhouse, H., Fu, D.J., de Pennington, N., Higham, A., Tham, Y.C., Jia, Y. and Habli, I. (2026). Building a code of conduct for AI-driven clinical consultations. Nature Medicine, [online] 32(2), pp.400–403. doi:https://doi.org/10.1038/s41591-025-04068-w.