Issue 3 · 9 May 2026 · Fortnightly | AI in Healthcare · Australia & Beyond
New peer-reviewed evidence on AI scribes lands this fortnight: the notes are mostly excellent — but the small fraction that aren't could seriously harm patients. Meanwhile, PwC and UTS have surveyed 133 Australian health leaders to find the real barriers to AI adoption, and your allied health digital software deadline just got a lot more real.
Here's what you need to know:
Summary
Read time: ~5 min
Top Stories
- The 5.3% Problem: AI Scribe Notes Are Mostly Excellent — But the Errors That Slip Through Can Seriously Harm Patients
- PwC + UTS: 62% of Australian Health Leaders Say Trust Is the Real AI Barrier — Not Technology
- Australia's Allied Health Digital Deadline Is July 2026 — Is Your Software Ready?
- The NDIS Commission Publishes Its AI Playbook — and Sets the Governance Standard for the Sector
Signals
- ACSQHC publishes AI Safety Scenario for ambient scribes — practical clinician framework
- My Health Record Sharing by Default Act now in effect — pathology and imaging first
- ADHA onboarding allied health vendors for My Health Record conformance ahead of July deadline
- Nature Medicine: LLMs match specialist diagnostic accuracy — but only with complete, structured clinical data
Top Stories
The 5.3% Problem: AI Scribe Notes Are Mostly Excellent — But the Errors That Slip Through Can Seriously Harm Patients
🌏 Global | Clinical Safety · ★ 7,545 AI-generated notes assessed · JMIR Medical Informatics
A landmark quality improvement study published in JMIR Medical Informatics has given the healthcare sector its most granular look yet at the safety profile of ambient AI scribes. Across a two-month pilot at UC Davis Health, 31 physicians used an AI scribe across 7,545 clinic visits — and then evaluated a random sample of their notes for errors. The headline finding: **94.7%** of AI-generated notes were free from significant errors.
But the remaining **5.3% is the story. Those notes contained errors rated as posing serious or imminent risk of patient harm if left uncorrected. The most common error type was accidental omission, present in 18% of evaluated notes — information the patient said that the AI simply failed to capture. Hallucinations appeared in 11.5%** of notes. Crucially, even when errors occurred, most were mild to moderate in severity. Only 19 notes across the entire pilot carried the highest risk ratings.
The researchers' conclusion: AI scribes are ready for clinical deployment, but they are not ready to be left unsupervised. The study provides a scalable model for pre-deployment safety piloting — evaluate a sample of notes before widespread rollout, establish ongoing monitoring, and treat clinician review as non-negotiable.
What it means for clinicians:
- **94.7%** accuracy is impressive — but in a busy practice seeing 20 patients a day, that's roughly one note per day that may need meaningful correction.
- Omissions (**18%) are more common than hallucinations (11.5%**) — the AI is more likely to miss something than to invent it. Review for completeness, not just accuracy.
- Before deploying any AI scribe, run a structured pilot: evaluate a sample of notes against a safety checklist, just as this study modelled.
PwC + UTS: 62% of Australian Health Leaders Say Trust Is the Real AI Barrier — Not Technology
🇦🇺 Australia | Industry Report · ★ 133 health leaders · 5 national roundtables · May 2026
PwC Australia and the University of Technology Sydney have published findings from five national roundtables with 133 Australian health and wellbeing sector leaders — the most comprehensive senior-level survey of Australian healthcare AI sentiment this year. The headline: AI adoption is growing, but it is being held back not by technology limitations but by human factors — principally trust, governance, and workforce readiness.
**62% of Australian health leaders identified trust and governance as their primary inhibitor to AI readiness. 49% pointed to traditional care and business models as a barrier. Yet the evidence of impact is accumulating: 53%** of healthcare CEOs say AI has already made their employees more efficient. The report documents a rapidly growing portfolio of Australian AI use cases — from AI-enhanced breast cancer detection at St Vincent's Melbourne to ear disease screening for Aboriginal and Torres Strait Islander children via DrumBeat.ai.
PwC's recommended path: establish governance first, define strategic AI ambition, build technological readiness, then invest in workforce preparation. The report warns against deploying complex clinical AI before low-risk administrative applications have built organisational confidence and skills.
What it means for clinicians:
- Trust and governance lead adoption barriers — not cost, not capability. Clinicians who want to use AI tools should start by establishing clear consent, documentation, and review workflows.
- The **53%** efficiency finding confirms that AI is delivering in practice — but only for organisations that prepared their workforce first.
- Allied health practices are not named in the roundtables, but the governance principles apply directly: start with administrative AI, build confidence, then move to clinical documentation tools.
Australia's Allied Health Digital Deadline Is July 2026 — Is Your Software Ready?
🇦🇺 Australia | Allied Health · ★ ADHA National Allied Health Digital Uplift Plan · Deadline: July 2026
The Australian Digital Health Agency has published its National Allied Health Digital Uplift Plan — a multi-year roadmap for bringing Australia's allied health workforce into the digital health ecosystem. Its Horizon 1 goal is concrete and time-bound: allied health clinical information systems must become My Health Record-conformant by July 2026.
ADHA is actively partnering with select software vendors to deliver conformant clinical information systems by that deadline, backed by communication strategies and adoption resources. Conformant systems give allied health professionals secure access to My Health Record — meaning patient records, referral letters, diagnostic results, and discharge summaries are accessible in the clinical workflow.
The plan also flags a longer-term priority: improving data integration between health and disability systems — a direct reference to the NDIS and DVA pathways. AI tools are explicitly identified across all three planning horizons, with a focus on safe deployment, data literacy, and equity. For allied health professionals yet to digitise their workflows, July 2026 is the clearest signal that the transition is no longer optional.
What it means for clinicians:
- If your clinical software is not yet My Health Record-conformant, contact your vendor now — the July 2026 deadline is weeks away.
- Conformant systems unlock real-time access to patient records at the point of care, reducing time spent chasing referrals and discharge summaries.
- AI tools are a named component of this roadmap — the ADHA is building toward a future where AI and My Health Record integration are standard practice for allied health.
The NDIS Commission Publishes Its AI Playbook — and Sets the Governance Standard for the Sector
🇦🇺 Australia | NDIS Regulation · ★ NDIS Quality and Safeguards Commission · March 2026
The NDIS Quality and Safeguards Commission has published a formal AI Transparency Statement, becoming one of the first Australian government agencies to publicly document exactly how it uses AI, what oversight is in place, and what principles govern its systems. The statement applies the OECD definition of AI and commits to the Digital Transformation Agency's mandatory requirements for responsible AI — including strategic oversight, operational preparedness, and impact assessment.
The Commission's definition is deliberately broad: systems that "analyse data, generate insights, and support decision-making in health and welfare contexts" — language that explicitly covers the AI documentation and outcome-tracking tools increasingly used by NDIS support providers and allied health practitioners. The statement commits to a use-case register, staff training protocols, and ongoing human oversight.
For NDIS practitioners who submit plans, reports, and progress notes into NDIS systems, this matters directly. The Commission is now applying the same AI governance lens to its own operations that it will increasingly expect from the providers it regulates. As automated quality-checking tools expand across the NDIS, practitioners with AI-assisted workflows will need to demonstrate equivalent accountability.
What it means for clinicians:
- The Commission's governance framework signals what it will expect from providers: documented oversight, human accountability, and explainability for any AI-assisted decision.
- NDIS practitioners using AI scribes, report generators, or outcome tools should maintain a log of which AI tools they use and how they review outputs — this is your accountability trail.
- The Commission's definition is broad enough to cover clinical documentation tools. Treat your AI vendor relationships with the same diligence you would any regulated supplier.
Signals
AU | Patient Safety — The Australian Commission on Safety and Quality in Health Care has published a dedicated AI Safety Scenario for ambient scribes — a practical, clinician-oriented framework covering consent, review workflows, error identification, and escalation pathways. Required reading before deploying any AI scribe in a clinical setting. Read →
AU | Legislation — The Health Legislation Amendment (Modernising My Health Record – Sharing by Default) Act 2025 is now in effect. Pathology and diagnostic imaging providers are the first required to upload results by default — allied health is next on the timeline. Patients will no longer need to manually opt in to share records. Read →
AU | Allied Health — ADHA is actively onboarding allied health software vendors to My Health Record conformance ahead of the July 2026 deadline. If your clinical information system vendor has not reached out about conformance, contact them directly — non-conformant software will leave you outside the national digital health ecosystem. Read →
🌏 Global | Research — A new Nature Medicine meta-analysis shows large language models match or exceed specialist physicians in diagnostic accuracy across 17 specialties — but only when structured clinical information is provided as context. Incomplete or unstructured patient data significantly degrades performance, reinforcing the importance of thorough clinical documentation. Read →