It’s been a year since I wrote about healthcare AI trends for 2025. Time for an honest assessment: where does healthcare AI in Australia actually stand in early 2026?

The short answer: progress is real but uneven. Hype has tempered. Hard work continues.

What’s Happened in the Past Year

Radiology AI has normalised in some settings. Major public health networks and large private radiology groups are running radiology AI in production. Chest X-ray triage and mammography AI are the most common. This isn’t universal—many smaller and regional sites haven’t implemented—but it’s no longer experimental in leading organisations.

Ambient documentation has landed. Several health services and practice groups have deployed AI documentation tools. Microsoft’s DAX is the most visible, but alternatives are appearing. Early feedback is positive on time savings, though accuracy concerns remain.

Governance has matured. More organisations have formal AI governance structures. The conversation has shifted from “should we govern AI?” to “how do we govern it effectively?” This is progress.

Investment has stabilised. The venture capital surge into healthcare AI has cooled. Health service AI budgets haven’t collapsed, but the easy funding of 2024-25 isn’t available. Projects need clearer business cases now.

A few notable incidents. Without naming specifics, there have been AI-related clinical incidents that made internal news even if not public headlines. These have prompted governance strengthening and more cautious adoption in affected organisations.

Where We Are by Sector

Public hospitals (metropolitan). Mixed adoption. Leading health networks have multiple AI systems deployed. Others are still in evaluation. The variability is striking—organisations with similar resources are at very different stages.

Public hospitals (regional). Mostly lagging. Infrastructure challenges, workforce limitations, and smaller volumes make AI implementation harder. Some shared services initiatives are emerging, but regional disparity is widening.

Private hospitals. Growing interest, driven by competitive positioning and efficiency pressure. AI is appearing in diagnostics, documentation, and operational optimisation.

Pathology. Slower than radiology. Digital pathology adoption is a prerequisite that many labs still don’t have. Where digitalisation exists, AI is following.

General practice. Limited adoption. Practice management AI and documentation tools are emerging, but diagnostic AI in primary care is minimal. Fragmentation and incentive structures are barriers.

Aged care. Early stages. Falls prediction and medication management AI are appearing in some residential facilities, but most aged care providers haven’t started.

The Maturity Spectrum

I’d characterise Australian healthcare AI maturity roughly as:

Leaders (5-10% of sector). Production AI deployments. Formal governance. Internal capability. Measuring outcomes. Planning next waves.

Early adopters (15-20%). Active pilots or early implementations. Governance developing. Building capability. Still learning.

Considering (30-40%). Aware of AI. Evaluating options. Haven’t committed. May have governance gaps.

Not engaged (30-40%). AI not on current agenda. Focused on other priorities. May lack awareness or capability.

This distribution varies by sector. Radiology is more advanced; primary care is less. Metropolitan is ahead of regional.

What’s Still Hard

Several challenges persist:

Integration complexity. Connecting AI to clinical systems remains expensive and time-consuming. Interoperability standards help but don’t eliminate the work.

Change management. Getting clinicians to adopt and trust AI is as challenging as the technology itself. Resistance, indifference, and inappropriate use patterns are common.

Evidence gaps. For many AI applications, local validation is limited. We’re relying on international evidence that may not translate perfectly.

Governance overhead. Doing governance well requires resources that compete with other priorities. Some organisations are under-governing; others are creating bureaucracy without value.

Vendor relationships. The healthcare AI vendor market is still maturing. Products disappear, vendors pivot, support quality varies.

Workforce constraints. Not enough people with health informatics and AI expertise. Competition for talent is intense.

What’s Working

Despite challenges, some things are working:

Clinical leadership. Where clinicians lead AI initiatives (not just endorse them), adoption succeeds. Clinical ownership remains the strongest predictor.

Focused applications. Organisations that start with specific, bounded use cases outperform those attempting broad transformation.

Governance integration. Embedding AI governance in existing clinical governance structures works better than creating separate silos.

Realistic expectations. Teams that expected incremental improvement have generally achieved it. Teams that expected transformation have been disappointed.

Learning networks. Informal networks where health informaticists share implementation experiences are accelerating learning across the sector.

What I’d Do Differently

Looking back at advice I gave organisations last year:

I underestimated change management needs. Technology implementation is maybe 30% of effort; culture and workflow change is 70%. I knew this intellectually but still underestimated in planning.

I was too optimistic on timelines. Implementations consistently took 50-100% longer than planned. I should have built more buffer.

I didn’t emphasise data quality enough. Organisations with poor underlying data struggled regardless of AI quality. Data remediation should happen earlier.

Looking Forward

For the rest of 2026, I expect:

Consolidation over expansion. Organisations will focus on making existing AI work better rather than adding new AI systems.

Governance tightening. Increased regulatory and accreditation attention on AI governance. Organisations without adequate governance will face pressure.

Vendor shakeout. Some healthcare AI vendors won’t survive the funding tightening. Due diligence on vendor viability matters.

Outcome evidence accumulating. As implementations mature, we’ll have better Australian data on what actually improves patient outcomes.

Workforce development. More attention to building health informatics and AI capability in the healthcare workforce.

My Advice for 2026

If you’re leading healthcare AI initiatives:

If you haven’t started: Don’t panic, but don’t ignore. Start building capability even if you’re not implementing. Understand what AI can do. Build governance foundations. Identify potential applications.

If you’re early in the journey: Focus on doing your current initiative well rather than adding more. Measure outcomes. Learn from experience. Build organisational capability.

If you’re more advanced: Share your learnings. The sector benefits when leading organisations help others avoid mistakes. Consider research partnerships to build the evidence base.

For everyone: Maintain realistic expectations. AI is useful but not transformational. Patient safety remains paramount. Governance isn’t optional.

We’re not at the beginning of healthcare AI in Australia, but we’re not at the end either. The hard, unglamorous work of making AI actually improve healthcare continues.

---

Dr. Rebecca Liu is a health informatics specialist and former Chief Clinical Information Officer. She advises healthcare organisations on clinical AI strategy and implementation.