Artificial intelligence is no longer a futuristic idea in medicine. From triage algorithms and imaging diagnostics to predicting deterioration before symptoms appear, AI now touches nearly every aspect of clinical care. Yet as clinicians learn to integrate these tools into practice, the law is quietly recalibrating what it means to deliver reasonable care.

The adoption of AI promises safer, faster, and more consistent decisions, yet it also introduces new dilemmas in ethics, accountability, and liability. The greatest medico-legal risk may soon lie not in how AI is used, but in failing to use it appropriately when evidence supports it.

From Optional to Expected

A landmark experiment published in the Journal of Nuclear Medicine asked 2,000 mock jurors to assess whether doctors were negligent when they either followed or ignored an AI recommendation. Jurors were just as likely to excuse a physician who acted on AI advice as one who followed traditional clinical judgment – that is even when the algorithm’s guidance later proved wrong.

That subtle shift in public perception hints at a deeper legal one: AI-assisted medicine is edging toward the new baseline. If jurors already view AI-guided decisions as “reasonable,” it is not difficult to imagine future courts treating validated AI systems as part of the accepted standard of care.

The Shifting Duty of Care

Questions of responsibility no longer stop at the clinician. A 2025 analysis by Babic et al. in npj Digital Medicine reviewed almost 1,000 FDA-cleared AI/ML medical devices and found most lacked meaningful follow-up data on performance once deployed. Existing oversight was built for static hardware, not adaptive software that continues to learn. The authors called for continuous “total-product-lifecycle” monitoring to ensure algorithms remain accurate and fair.

Until such adverse reporting systems for AI/ML mature, courts and professional bodies may define what counts as “safe and reasonable” use. If validated tools demonstrably improve accuracy, ignoring them could appear negligent; yet blind reliance on untested or drifting models could be equally risky. For now, the legal system is being asked to fill the gaps left by incomplete regulation.

Access and Equity

Liability also depends on access. A 2025 npj Digital Medicine study by Burns et al. found that locally built, task-specific machine-learning models outperformed large generative systems such as ChatGPT-4 in both accuracy and reliability, but were costly to develop and maintain. Many hospitals, especially smaller or safety-net providers, simply cannot afford them.

While those disparities persist, courts are unlikely to treat non-use of AI as negligence. But, as prices fall and validated systems become standardised across the NHS, economic defences will fade and the duty to adopt AI may solidify into professional expectation.

Trust and Transparency

As emerging analyses have warned, increasingly autonomous or “agentic” AI systems may blur traditional boundaries of accountability in healthcare. Hospitals must therefore audit the quality of their underlying data and remain transparent about how algorithms influence clinical decisions. Explainability, documentation, and informed consent will determine whether AI earns its place as a legitimate partner in patient care. 

The New ‘Reasonable’ Clinician

Taken together, these developments mark a turning point for both expert witnesses and the courts. Future negligence claims may no longer hinge on a doctor’s personal judgment, but on whether they engaged appropriately with available AI tools. The “reasonable clinician” of the next decade will be expected to understand, question and document how AI contributes to care. Neither deferring blindly to machines nor dismissing them outright.

As AI becomes the new colleague in the consulting room, the definition of professional responsibility is evolving. The next generation of cases may not ask whether a doctor made an error, but whether they used the intelligence at their disposal. In that sense, the standard of care itself is learning – and so must we.

By Jaya McCorry

References

Babic, B., Glenn Cohen, I., Stern, A.D. and others (2025) ‘A general framework for governing marketed AI/ML medical devices’, npj Digital Medicine, 8, 328. Available at: https://doi.org/10.1038/s41746-025-01717-9

Burns, M.L., Chen, S.Y., Tsai, C.A. and others (2025) ‘Generative AI costs in large healthcare systems: an example in revenue cycle’, npj Digital Medicine, 8, 579. Available at: https://doi.org/10.1038/s41746-025-01971-x

Cohen, I.G., Babic, B., Gerke, S. and others (2023) ‘How AI can learn from the law: putting humans in the loop only on appeal’, npj Digital Medicine, 6, 160. Available at: https://doi.org/10.1038/s41746-023-00906-8

Gerke, S., Babic, B., Evgeniou, T. and Cohen, I.G. (2020) ‘The need for a system view to regulate artificial intelligence/machine learning-based software as medical device’, npj Digital Medicine, 3, 53. Available at: https://doi.org/10.1038/s41746-020-0262-2

Pham, T. (2025) ‘Ethical and legal considerations in healthcare AI: innovation and policy for safe and fair use’, Royal Society Open Science, 12(5), p. 241873. Available at: https://doi.org/10.1098/rsos.241873

Price, W.N. II, Gerke, S. and Cohen, I.G. (2021) ‘How much can potential jurors tell us about liability for medical artificial intelligence?’, Journal of Nuclear Medicine, 62(1). Available at: https://doi.org/10.2967/jnumed.120.257196