Artificial intelligence (AI) is already entering clinical practice and patient-facing care.

Used well, it could reduce paperwork, improve screening and diagnosis, support earlier intervention, and help the health system make better use of scarce specialist capacity. Used poorly, it could exacerbate existing barriers to care, make responsibility for errors unclear, and weaken the human relationships that good care depends on.

In Brave New Ward? Towards a national pathway for medical AI, Researcher Debora Spiekermann examines how New Zealand currently governs medical AI, how its adoption may affect patient care, and where it could offer the greatest benefit. The paper also considers what New Zealand can learn from Singapore’s more coordinated governance approach.

The paper argues that New Zealand should adopt medical AI selectively and with ambition. Adoption should follow a clear national pathway, with enforceable standards, validation for New Zealand populations, ongoing monitoring, clear responsibility, and a commitment to person-centred care.

When adoption outpaces governance

New Zealand has no AI-specific legislation and no single regulator responsible for medical AI. Existing health law, professional obligations, institutional arrangements, and non-binding guidance all play a part, but they do not form a comprehensive statutory regime.

There are several reasons that good governance is essential. Medical AI systems can change after deployment due to updates, learning, and adapting. They require regular auditing for accuracy, bias, and performance drift. A tool trained overseas may also work differently for New Zealand populations, particularly Māori, Pacific people, and other groups underrepresented in international datasets.

The risks are not limited to technical challenges. If AI tools  overemphasise efficiency, the measures used to evaluate them may give too much weight to patient throughput. An AI scribe might give a doctor more time to listen to a patient, but it could also create an expectation that the doctor sees more patients. That is a policy choice, not an inevitable result of the technology.

What the paper argues

The paper brings together three areas of analysis:

Lifecycle governance: how medical AI should be governed from procurement and validation through clinical use, monitoring, and withdrawal

Models of health and care: how different understandings of health shape the use of AI in patient care, including the biomedical model and person-centred care

Priority applications: where medical AI may offer the greatest value, such as diagnostic and screening support, targeted prevention, and administrative automation

Its central argument is that New Zealand should adopt medical AI selectively and with ambition, focusing on high-value uses that meet urgent local needs. Those uses should be governed across their whole lifecycle, with a clear public purpose and a commitment to person-centred care. Because applications carry different risks, governance should be risk- and function-based.

Key findings

New Zealand’s governance arrangements are flexible but not yet coherent. There is no complete, publicly accessible overview of which systems are in place or under evaluation, and responsibility for ongoing monitoring remains diffuse.

The strongest opportunities are targeted: diagnostic and screening support, targeted prevention, and administrative automation. But even seemingly low-risk tools, such as AI scribes, raise questions about consent, data security, and the dynamics of the doctor-patient relationship.

Safe adoption depends on stronger digital foundations, good data, validation for New Zealand populations, ongoing monitoring, and clinician responsibility. Success should be measured not only by throughput, but also by access to specialist care, administrative burden, time with patients, clinician-patient relationships, and staff burnout.

Recommendations for New Zealand

The paper makes 10 recommendations:

• Establish a public national register of medical AI systems
• Make AI guidelines enforceable through a named authority
• Require pre-deployment validation for New Zealand populations
• Require ongoing blind evaluation against human controls
• Require specific liability and clear lines of responsibility during procurement and clinical use
• Require data portability, interoperability, and withdrawal mechanisms
• Invest in the foundations for safe adoption
• Promote responsible innovation in medical AI
• Make AI adoption values-led
• Keep healthcare person-centred