Introduction by Croakey: The increasing importance of artificial intelligence (AI) as a powerful and pervasive determinant of health is not sufficiently recognised at this critical time for policy development, according to experts in the field.
Dr Raffaele F. Ciriello, a Senior Lecturer in Business Information Systems at the University of Sydney Business School, and Professor Kathryn Backholer, Professor of Public Health Policy at Deakin University, call for public health expertise and institutions to be centred in policy responses to the AI juggernaut.
“If AI increasingly shapes the information people see, the choices they make, the support they receive and the commercial pressures they face, then consequential AI systems should increasingly be governed as public interest infrastructure,” they write below.
Raffaele F Ciriello and Kathryn Backholer write:
Australia’s AI debate is intensifying, with wide-ranging implications for health and wellbeing. Governments and communities argue over data centres, electricity, water, copyright, investment and sovereign capability. The National AI Plan is framed around making the economy more competitive, productive and resilient, with the broader policy agenda focused heavily on infrastructure, capability, investment and safety.
That agenda is moving: National Cabinet has now agreed to develop nationally consistent mandatory standards for large data centres covering energy, water and land use, with legislation intended for early 2027.
These are important developments, but Australia’s policy debate still largely treats health as a sector in which AI can be applied, rather than an outcome shaped by AI across everyday life.
Public health brings a distinct perspective. It has traditionally looked upstream. Health is shaped not only by what happens in hospitals and clinics, but by the conditions in which people are born, grow, live, work, and age. AI is now changing many of those conditions.
World Health Organization-backed research has identified digitalisation as reshaping social, commercial, political and economic determinants of health. General purpose AI takes this a step further. It is becoming part of the everyday environment through which established determinants operate.
Consider commercial influence. Conversational AI is moving advertising from search results into private exchanges where people discuss diet, stress, debt, relationships, parenting and health.
In our recent article in The Lancet, ‘Targeted advertising in generative artificial intelligence chatbots: a new public health risk’, we argued that this combination of intimacy, contextual knowledge, and commercial incentives could make targeted persuasion especially consequential during moments of vulnerability.
The concern is not just a personalised advertisement, but a business model that can shape the information people receive in ways that serve commercial or political interests. Public health has seen this movie before with tobacco, alcohol, gambling and unhealthy food: voluntary safeguards rarely hold water when they conflict with commercial incentives.
Or consider information-seeking. People increasingly ask chatbots questions that are plainly health-relevant even when those systems are not regulated as health products.
A 2026 systematic review found that generative AI can increase the speed and scale at which convincing health misinformation is produced, while some studies found that users struggled to distinguish AI-generated misinformation from human-authored material.
Changing information environment
The public health challenge therefore extends beyond whether a chatbot gives an accurate answer to a clinical or other health-related question. It also concerns how AI changes the information environment in which everyday health decisions are made, and whether its benefits and harms are distributed unequally.
The same ambiguity appears in relationships and mental health. AI companions and general-purpose chatbots can provide immediate, personalised support where human care is out of reach. But open-ended commercial systems are not bounded therapeutic interventions.
A 2026 study in Nature Human Behaviour found that intensive and highly disclosive AI-companion use was associated with lower wellbeing, particularly among people with smaller offline social networks. A Nature Medicine study auditing multi-turn conversations found that apparently supportive chatbot responses could sometimes reinforce the psychological mechanisms underlying a simulated user’s vulnerability.
The sticky question is which users, interaction patterns and chatbot design features increase or decrease this risk, and whether existing safeguards actually improve health outcomes.
These examples point to a broader governance problem. We are getting better at measuring what AI systems do, but much less is known about what happens to people and populations as a result. In mental health, for example, companies may be able to report whether a chatbot recognised suicidal language, offered crisis support or blocked a harmful response.
But public health needs to know what happens afterwards. Did the person reach care? Did help-seeking increase or decline? Were some groups more likely to experience harm? Did a safeguard actually reduce self-harm or merely satisfy an internal safety benchmark?
Recent work on youth suicide risk reaches much the same conclusion: we remain strikingly short of real-world outcome evidence. At present, platform interaction data and population-health outcome data largely sit in separate worlds. Independent, privacy-preserving monitoring should connect them without turning intimate conversations into routine objects of corporate or state surveillance.
Equity runs through all of this. The benefits of AI will not be distributed evenly, nor will the burdens.
For example, subscription chatbot models can determine who gets an ad-free or higher-quality system. Language, geography, disability, digital literacy and socioeconomic advantage shape access and exposure.
AI has a physical footprint too. Australia’s data-centre boom is already producing disputes over electricity, water, land, noise and who pays for the infrastructure supporting it. These are also questions of equity: who benefits, who carries the costs, and where power becomes concentrated.
Calls to action
Australia therefore needs to bring a population health lens to AI governance.
Recent Croakey commentary has highlighted the same gap, calling for public health and health equity expertise in Australia’s emerging AI standards process. A related Croakey article argued that public health actors need to engage directly rather than watch from the sidelines.
The new Office of AI should formally involve public health and health equity expertise in developing national standards, working with health departments, the Australian Centre for Disease Control and other public health bodies. Public health organisations and researchers should, in turn, engage in consultations and standards-setting and build the evidence needed to identify emerging health effects.
The public health field itself also needs greater AI capability across teaching, research and practice, not only to use AI in healthcare, but to understand how it is reshaping the commercial and social determinants of health.
Researchers need meaningful access to data so health effects can be independently monitored rather than defined by company metrics. Governments should regulate harmful commercial practices across platforms rather than leaving companies to police themselves.
Australia needs enough independent expertise and infrastructure to avoid becoming wholly dependent on the same companies it is trying to regulate. The National AI Plan already recognises this logic for secure public-sector AI infrastructure; it could be the beginning of a more ambitious public interest capability.
If AI increasingly shapes the information people see, the choices they make, the support they receive and the commercial pressures they face, then consequential AI systems should increasingly be governed as public interest infrastructure.
This does not mean government must build every model or nationalise the entire AI stack. It means public institutions such as the Office of AI, the Australian CDC, the eSafety Commissioner, and health departments need enough capacity, transparency and leverage to protect population wellbeing when commercial interests point in another direction.
The window to build that capacity is now, while the infrastructures, norms and revenue models are still being formed.
Author details
Dr Raffaele F Ciriello is a Senior Lecturer in Business Information Systems at the University of Sydney Business School. His research examines compassionate digital innovation, AI companions, and the governance of emerging technologies for the common good. He is a voluntary member of the eSafety Commissioner’s Parent Advisory Group. Kathryn Backholer is Vice President (Policy) of the Public Health Association of Australia.
Professor Kathryn Backholer is Professor of Public Health Policy at Deakin University and Co-Director of the Global Centre for Preventive Health and Nutrition. Her research examines the commercial determinants of health, digital environments, and policy approaches to improve population health and equity. She is Vice President (Policy) of the Public Health Association of Australia.
Generative AI Disclosure: We used ChatGPT to copy-edit our own text for clarity and concision. We manually validated and corrected all output. We retain ownership and responsibility for all intellectual contributions.
Also read: As Federal Government moves to address online safety, what do health experts think?
See Croakey’s archive of articles on AI and health
Add Comment