Artificial intelligence is already reading chest X-rays, screening for diabetic retinopathy, flagging anomalies in pathology reports, and generating clinical notes from consultations. The adoption curve is steep. A 2024 survey published in JMIR Mental Health found that 43% of Australian mental health professionals are already using AI tools in their practice. In radiology and pathology, uptake is higher still.
The clinical case for these tools is understandably strong. AI diagnostic systems have demonstrated detection rates that match or exceed specialist performance on specific, narrow tasks. But the legal framework, by necessity, is playing catch-up. In the event of a catastrophic misdiagnosis, injured patients and their lawyers will find themselves in genuinely uncharted territory.
The central problem is one of attribution of responsibility. When a human clinician misses a tumour on a scan, the negligence framework is relatively settled: we ask whether the doctor’s conduct fell below the standard of a reasonably competent practitioner, applying the principles the High Court has articulated in cases such as Rogers v Whitaker [1992] HCA 58. But what is the reference point when the clinician relied on an AI tool to perform the primary analysis, and the AI was wrong?
The practitioner remains responsible - at least for now
AHPRA’s guidance on professional obligations when using AI in healthcare, published in August 2024, is unambiguous: regardless of what technology a practitioner uses, the responsibility for safe and quality care remains theirs. The guidance makes clear that AI is a tool, not a co-practitioner. That is in keeping with the general state of the law as it currently exists. If a clinician delegates their judgment to an algorithm without adequate independent review, they have not discharged their duty of care, they have outsourced it.
Even the various protective legislation introduced by States & Territores, which seeks to aid the medical profession’s defence of claims, does little to assist. The Civil Liability Acts across Australian states and territories provide a defence where a practitioner’s conduct is consistent with widely accepted peer professional practice. But that defence is already under strain in the AI context: when the technology is new, peer practice is still forming, and no settled standard has yet emerged. A practitioner cannot point to what colleagues are doing when what colleagues are doing has not yet been established.
The legal approach will inevitably evolve over time as the technology and adoption rates change. Perhaps a time will come when, with certain medical services, the community expectation is that the AI can and should be relied upon independent of clinician involvement. But that time seems a long way off.
What About Informed Consent?
The issue of informed consent is also a critical issue. Rogers v Whitaker established that patients have the right to know about material risks inherent in proposed treatment. It’s about whether a reasonable patient (and not the doctor) would attach significance to it. A plausible argument exists that patients should be informed when AI is involved in their diagnosis or treatment planning, particularly where that AI carries a known error rate in the relevant presentation type. Failure to disclose may, in the right circumstances, ground a separate basis for liability.
The need for disclosure of AI involvement in the legal setting is well known already. The judiciary throughout the country is already calling for transparency about its use and are consistently reinforcing the need for practitioners to exercise an overriding independent judgment themselves. The various Law Societies are also emphasising the need for disclosure of the use of AI to clients.
Why then would the medical profession be any different? At this relatively early stage of AI adoption (if not beyond) people have a right to know who or what the true source of the opinion or recommendation is. The use of AI in medical services really ought to be made known to patients to help them assess all material risks in their treatment regimes.
Doctor or AI: Who to Sue?
If the AI gets it wrong, in practice, injured patients and their lawyers will likely pursue both avenues. The healthcare provider will be sued because they are insured, identifiable, and in a direct relationship with the patient. The AI developer will be joined because if the AI system itself was defective, if its training data was skewed, or its validation inadequate for the specific patient population, then manufacturer liability is properly engaged. These are not mutually exclusive claims.
The causation challenge
Perhaps the most difficult issue for injured plaintiffs will be causation. Establishing that the AI’s error caused the harm (rather than the clinician’s independent decision, or the underlying condition itself) will require expert evidence of a kind courts have not yet had to evaluate. You can’t cross-examine a neural network. How do you establish that a different output from the algorithm would have led to a different clinical decision, and a different outcome for the patient?
These and many other questions will need to be answered. They are the questions that will define how Australian courts approach this emerging liability landscape, and they will likely be asked sooner than the profession expects.
What Medical Professionals need to know
While the ongoing and regular use of AI is now entrenched and has many patient and consumer benefits there is still a balancing act to be had. In keeping with the longstanding principles of being a professional, matters of integrity, morality and altruism must remain paramount.
It will rarely be sufficient for clinicians to simply defer to the judgment of AI, no matter what the workload or external pressures are. The law will demand the individual care and attention of the clinician. They are, in effect, the captain of the medical ship.
What injured patients need to know
For patients who believe they have been harmed by a missed or wrong diagnosis, particularly where AI systems were involved in their care, the legal landscape is complex but remains navigable. Claims may lie against the treating clinician, the healthcare facility, the AI developer, or some combination of all three. The threshold question of whether AI was used, and in what capacity, is now a legitimate and important line of inquiry in any medical negligence matter.
The law in this area is, for now, unwritten. The first significant Australian decision on AI-assisted misdiagnosis will set the preliminary framework for everything that follows. Patients harmed in the meantime should not assume the absence of decided cases means the absence of a claim. It means, instead, that the ground is shifting and that specialist advice has never been more important.
