AI is a tool, and using it doesn’t transfer clinical responsibility

Agreement: I Agree Body: Dear Editor, Daniel Sokol’s opinion piece on doctors using AI raises sound arguments [1]. It is true that AI is a tool and using it doesn’t transfer the clinician’s responsibility to the tool. Doctors need to still be criticized for the decisions they make with or without AI assistance. But increasingly, AI is operating autonomously. That changes the picture. New questions arise. What boundaries does it operate within? Who monitors it? And who is accountable when something goes wrong? Take triage, for example [2, 3]. An AI system might independently assign a patient to a care pathway, schedule or delay a review, and decide whether a clinician is needed. Another example is patient follow-up [4]. An AI system can check whether a patient is progressing as expected, and flag when a clinician needs to step in. In these situations, the system is exercising bounded autonomy. AI systems might be trained on data from real clinical scenarios to make these decisions. But is that enough? Could there be outliers the system has not learned to recognise? Because available datasets are not necessarily the gold standard, we may not know what is missing. Plus, data alone may not capture how care works on the ground. Clinicians can help identify those gaps and work with technical developers to design tools that fit routine practice, followed by proper validation and safety checks. Ultimately, “human warranty” is essential [5]. Yet regardless of how the system was trained or who developed it, if it makes a decision that affects a patient, who should be held responsible? The designer? The clinician or hospital that chose to use it? Or the wider system that allowed it? These questions are still being worked through, and we need open discussions about them, while legal liability remains context dependent. References [1] Sokol D. Six rules for doctors using AI. BMJ 2026; 394: e100883. [2] Min C, Lim RXC, Tan SW, et al. Experience developing a pediatric medical chatbot in Singapore: a digital innovation for improved emergency care. Front Digit Health 2025; 7: 1557804. [3] Wempe M, Holleman F, Schinkel M, et al. Concordance between an artificial intelligence self-triage programme and physical triage. Emerg Med J 2026; 43: 606–611. [4] Meinert E, Milne-Ives M, Lim E, et al. Accuracy and safety of an autonomous artificial intelligence clinical assistant conducting telemedicine follow-up assessment for cataract surgery. EClinicalMedicine 2024; 73: 102692. [5] Ethics and governance of artificial intelligence for health. World Health Organization (2021). https://www.who.int/publications/i/item/9789240029200 (accessed 16 October 2026). No competing Interests: Yes The following competing Interests: Electronic Publication Date: Friday, October 16, 2026 - 07:02 AI use: Yes I have used AI AI use details: ChatGPT was used to check grammar and spelling Highwire Comment Subject: Workflow State: Released Full Title: AI is a tool, and using it doesn’t transfer clinical responsibility Highwire Comment Response to: Check this box if you would like your letter to appear anonymously:: Last Name: Kaliya-Perumal First name and middle initial: Arun-Kumar Email: arunkumar.kp@ntu.edu.sg Address: R3 Institute Singapore, Nanyang Technological University, Singapore Occupation: Clinician-Scientist Affiliation: BMJ: Additional Article Info: Rapid response