Australia likes to think of itself as a fast adopter of new technology, but when it comes to putting guardrails around artificial intelligence the country keeps arriving late to its own party. That tension sits at the heart of a new argument, published by MobiHealthNews, that Australia has an AI governance problem it has yet to resolve. The point lands hardest in healthcare, a sector racing to embed AI into diagnosis, triage and administration while the rules meant to keep patients safe remain a work in progress.
The context is familiar to anyone who has followed the policy debate. Australia does not lack for principles. The federal government published a set of AI ethics principles back in 2019, and in September 2024 it released a Voluntary AI Safety Standard alongside a proposals paper floating mandatory guardrails for AI in high-risk settings. What it has lacked is teeth. Most of the existing framework is voluntary, and the promised mandatory rules for high-risk uses have moved slowly through consultation rather than into legislation. The result is a patchwork that leaves developers, hospitals and clinicians guessing about where the lines actually sit.
Why health is the sharp end
Healthcare is where the abstract debate about AI governance becomes concrete. A poorly calibrated model that recommends the wrong dose, misreads a scan or quietly discriminates against a subgroup of patients does not just dent a balance sheet. It can hurt someone. That is why the sector already carries more regulation than most, and why the gaps feel more dangerous here than in, say, marketing or logistics.
Software that makes clinical claims can fall under the Therapeutic Goods Administration as a medical device, and the TGA has been clarifying how its framework applies to AI-enabled and machine-learning products. But a great deal of the AI now creeping into Australian clinics does not announce itself as a medical device at all. Ambient scribing tools that draft consultation notes, chatbots that field patient questions, scheduling systems that prioritise appointments and large language models that summarise records often sit in a grey zone. They shape care without being regulated as though they do, and that is precisely the governance gap the critique is pointing at.
Two ways to read the problem
There are broadly two camps in this argument, and both have a case. The precautionary view, common among clinicians, privacy advocates and patient groups, holds that Australia is sleepwalking. On this reading, voluntary standards are no substitute for enforceable law, and the country risks importing tools built and trained overseas, on overseas populations, without the local scrutiny needed to know whether they are safe for Australian patients. Supporters of this position want mandatory guardrails for high-risk health AI, clear accountability when something goes wrong, and transparency about how models are trained and validated.
The competing view, more often heard from industry, health-tech founders and some hospital administrators, is that heavy-handed rules could smother genuinely useful innovation before it matures. Australia already struggles to commercialise its research, and a thicket of duplicated approvals could push developers to launch in friendlier markets first. This camp tends to favour risk-based, proportionate regulation that leans on existing bodies such as the TGA and the privacy regulator rather than a sprawling new AI act. The worry is not that governance is unnecessary, but that badly designed governance could deliver the worst of both worlds: slow to protect patients, yet fast to deter builders.
The uncomfortable truth is that these positions are not as far apart as they sound. Almost everyone agrees the current settings are too vague. The fight is over what replaces them, and how quickly.
The Australian stakes
For Australia specifically, the governance question is tangled up with sovereignty and trust. Much of the health system runs on stretched public budgets, and AI is being sold as a way to ease workforce shortages, cut administrative load and free clinicians for patient-facing work. That promise is real, and it is one reason adoption is accelerating even without settled rules. But a health system that deploys tools it cannot properly audit is taking on a risk that compounds quietly until something fails in public.
There is also a data dimension. Australian health data is sensitive, valuable and, under the Privacy Act, protected. Feeding it into AI systems, particularly models hosted offshore, raises questions the current framework answers only partially. A review of the Privacy Act has been grinding forward, and reforms could reshape how health data is used to train and run AI, but until those changes bed down, providers are making consequential decisions with incomplete guidance. Trust is the currency here, and Australians have shown before, most memorably during the My Health Record opt-out saga, that they will walk away from digital health if they feel their information is being handled carelessly.
What happens next
The near-term path runs through Canberra. The government has signalled it wants a risk-based approach, and the key decision is whether the promised mandatory guardrails for high-risk AI become law, and whether healthcare is explicitly named as a high-risk domain. If it is, hospitals, software vendors and clinicians will need to prove their systems meet a standard rather than simply assert that they do. The Albanese government’s broader productivity agenda has leaned on AI as a growth lever, which cuts both ways: it raises the political incentive to let the technology run, and the political cost of a high-profile failure.
Expect the pressure to keep building from professional bodies too. Medical colleges and nursing organisations have been issuing their own guidance on AI use, partly because official rules have not kept pace. That bottom-up governance is useful, but it is also a symptom of the vacuum. The longer formal rules take, the more the sector improvises, and the harder it becomes to bring everyone back to a single, coherent standard.
Resolving Australia’s AI governance problem, in the end, is less about choosing between innovation and safety than about being honest that the country has been trying to have both without paying for either. Healthcare is where that bill is likely to come due first, and where getting the settings right will matter most.
Sources: MobiHealthNews.
