When AI sets the standard, should accountability follow authority?

Agreement: I Agree Body: Dear Editor Morgan compellingly argues that, as artificial intelligence (AI) begins to outperform physicians in some cognitive tasks, the role of doctors may shift from being the best diagnosticians to becoming moral agents, system-level stewards, translators of evidence, and the “manual override” when technology fails.¹ Yet this proposed role raises a further question: can physicians remain fully accountable if they no longer retain meaningful decisional authority? This

BMJ

By AnonymousSeptember 16, 2026South Africa

Agreement: I AgreeBody:

Dear Editor

Morgan compellingly argues that, as artificial intelligence (AI) begins to outperform physicians in some cognitive tasks, the role of doctors may shift from being the best diagnosticians to becoming moral agents, system-level stewards, translators of evidence, and the “manual override” when technology fails.¹ Yet this proposed role raises a further question: can physicians remain fully accountable if they no longer retain meaningful decisional authority?

This tension becomes particularly important when AI performance moves beyond decision support and begins to influence institutional standards of care. Morgan cites evidence that human-AI combinations can perform worse than AI alone when AI already outperforms humans.¹ ² If such evidence leads hospitals to embed validated AI recommendations into clinical pathways, physicians may face a two-sided accountability problem. Following an erroneous AI recommendation may leave the clinician responsible for the resulting harm; overriding an AI recommendation that later proves correct may invite the opposite question—why did the clinician disregard a system known to outperform unaided physicians?

The result could be accountability without authority. A physician may remain nominally “in the loop” while institutional protocols, workflow design, performance metrics, and the evidential superiority of the AI substantially constrain the freedom to disagree. In such circumstances, human oversight risks becoming ceremonial rather than meaningful.

Recent work on medical AI oversight supports this distinction. Van de Sande and colleagues argue that meaningful oversight requires not merely human presence but epistemic capacity, cognitive space, decisional authority, and an effective ability to intervene.³ WHO guidance likewise emphasises accountability across the stakeholders who design, deploy, and use health AI, rather than treating responsibility as a burden borne by the end-user alone.⁴ Questions of liability arising from clinical AI have similarly been recognised as extending beyond simple clinician error.⁵

The governance challenge, therefore, is not solved merely by insisting on a “human in the loop.” Health systems should define, before deployment, who has authority to accept or override AI recommendations, under what circumstances disagreement is expected, how such disagreement should be documented and audited, and how responsibility is distributed among clinicians, healthcare institutions, developers, and deployers. Cases of clinician-AI disagreement should themselves become an important safety metric: not simply whether the human overrode the machine, but whether that override improved or worsened patient care.

Morgan concludes that the doctor may increasingly become the referee rather than the player.¹ But a referee can only be responsible for the game if they have genuine authority to enforce its rules. In AI-mediated medicine, responsibility should follow meaningful authority. Otherwise, retaining the physician as the final accountable human may preserve the appearance of human control while quietly transferring the substance of control elsewhere.

References

1.Morgan M. Matt Morgan: After checkmate-if AI is better, what are doctors for?. BMJ. 2026;394:e100753. doi:10.1136/bmj-2026-100753.
2.Vaccaro M, Almaatouq A, Malone T. When combinations of humans and AI are useful: A systematic review and meta-analysis. Nat Hum Behav. 2024;8(12):2293-2303. doi:10.1038/s41562-024-02024-1.
3.van de Sande D, Economou-Zavlanos N, van Genderen ME. Meaningful oversight of medical AI beyond human in the loop. NPJ Digit Med. 2026;9(1):569. doi:10.1038/s41746-026-02971-1.
4.World Health Organization. Ethics and governance of artificial intelligence for health: WHO guidance. Geneva: World Health Organization; 2021.
5.Mello MM, Guha N. Understanding Liability Risk from Using Health Care Artificial Intelligence Tools. N Engl J Med. 2024;390(3):271-278. doi:10.1056/NEJMhle2308901.

No competing Interests: YesThe following competing Interests: Electronic Publication Date: Wednesday, September 16, 2026 - 13:54AI use: No, I have not used AIHighwire Comment Subject: