Skip to main content
guide·11 min read

NSW Health AI Framework: What It Is and What It Means for Your Practice in 2026

The NSW Health AI Framework sets the rules for clinical AI adoption in practices. Here's what compliance looks like — and where automation fits safely in 2026.

Published 15 November 2026

NSW Health AI Framework: What It Is and What It Means for Your Practice in 2026

If you're running a medical practice, clinic, or health service in New South Wales right now, you've likely noticed that AI is no longer a futuristic talking point. It's sitting inside your practice management software, your inbox, and potentially your clinical decision workflow. The question is no longer whether AI is relevant to your practice — it's whether you're deploying it in a way that is clinically safe, legally defensible, and aligned with the governance standards your accreditor and regulator expect.

That's where the NSW Health AI Framework comes in.

In plain terms: This framework is a governance architecture — not a prohibition. For practices that understand it, it's a roadmap to confident, compliant AI adoption — not a barrier to it.

What Is the NSW Health AI Framework?

The NSW Health AI Framework is the foundational governance document that sets out how artificial intelligence should be adopted, deployed, and monitored across New South Wales health services. First published as part of NSW Health's broader digital health strategy in 2023 and updated progressively since, it addresses the full lifecycle of AI in healthcare — from procurement and validation through to post-deployment monitoring and incident response.

The framework applies directly to NSW Health facilities and district health services, but its principles are increasingly shaping expectations for all healthcare providers operating in New South Wales — including independent general practices, specialist clinics, allied health providers, and private hospitals. If you're seeing AI governance language appear in your accreditation standards or in correspondence from your Primary Health Network, this is why.

The NSW Health AI Framework operates alongside several other regulatory instruments that health practices need to understand:

  • The Australian Government's AI Ethics Framework, which establishes eight principles for responsible AI: human, societal and environmental wellbeing; human-centred values; fairness; privacy and security; reliability and safety; transparency and explainability; contestability; and accountability
  • TGA guidelines on Software as a Medical Device (SaMD), which classify AI tools that make or influence clinical decisions as regulated therapeutic goods requiring ARTG listing or appropriate exemption
  • The Privacy Act 1988 (Cth) and its 13 Australian Privacy Principles, which govern how patient data is collected, used, stored, and disclosed
  • My Health Record Act 2012, relevant wherever AI interacts with shared health record data
  • Australian Digital Health Agency standards, including FHIR interoperability requirements for clinical system integration

The NSW Health AI Framework doesn't replace any of these — it integrates them into a coherent governance layer specific to the NSW health system. Understanding the framework means understanding how these instruments interact.

The Six Core Principles

The NSW Health AI Framework is organised around six governing principles. Understanding these in operational terms — not just as abstract values — is what separates meaningful compliance from box-ticking.

1. Clinical Safety and Benefit

Any AI tool used in a clinical context must demonstrate clear patient benefit that outweighs its risks. This rules out a significant proportion of generic AI tools that haven't been validated in an Australian clinical population or in the specific clinical context in which they're being used. A tool that works well in a US hospital system is not automatically appropriate for an Australian general practice.

2. Governance and Accountability

Responsibility for AI decisions cannot be delegated to the tool itself. Clinicians and health service leaders retain accountability for all patient care decisions, even where AI has informed them. Every AI system in deployment must have an identifiable governance owner — someone who can answer for the tool's performance, decisions, and ongoing appropriateness.

3. Data Stewardship

Patient data used to train, run, or evaluate AI systems must comply with existing privacy legislation and NSW Health data governance policies. This has direct implications for cloud-hosted AI tools: if your vendor cannot tell you exactly where patient data is stored and who can access it, you have a compliance gap — and that gap sits with the practice, not the vendor.

4. Transparency and Explainability

Patients and clinicians should be able to understand, at an appropriate level, how an AI system reached a recommendation or output. Black-box systems that cannot be interrogated or audited are a governance red flag — and an increasingly active concern at the federal regulatory level, particularly following the Privacy Act reforms.

5. Equity and Non-Discrimination

AI systems must not reproduce or amplify existing biases that disadvantage particular patient populations. This has direct implications for the training datasets underlying any clinical AI tool you're considering. Asking a vendor where their training data came from is not a pedantic question — it's a governance requirement.

6. Continuous Monitoring

AI is not a set-and-forget deployment. The framework requires ongoing evaluation of AI system performance against its stated purpose — including post-deployment monitoring for performance drift, accuracy degradation, and changed use patterns. This is the principle most commonly overlooked by practices that have deployed AI tools and moved on.

What Does This Mean for Your Practice?

The practical implications of the NSW Health AI Framework depend on what kind of AI you're using and where in your practice it sits. The framework draws a meaningful distinction between clinical and administrative AI — and the two have very different compliance pathways.

Clinical Decision Support Tools

If your practice uses AI to assist with diagnosis, prescribing, care planning, triage, or risk stratification, you're in TGA territory as well as framework territory. Under TGA's SaMD classification guidance, software that uses patient data to generate recommendations that influence clinical decisions is likely a Class IIa or higher medical device — requiring appropriate registration or listing on the Australian Register of Therapeutic Goods (ARTG) before it can be legally used in patient care.

This is catching out practices that have adopted clinical AI tools from international vendors without verifying their regulatory status in Australia. In 2026, TGA enforcement in this space is more active than it was three years ago, and the argument that a tool is just decision support is not holding up under regulatory scrutiny. The classification question turns on whether the tool's output influences a clinical decision — not on how the vendor markets it.

Before deploying any clinical AI tool, your practice should be able to answer:

  • Is this tool TGA-registered or exempt under a valid basis?
  • Has it been validated in a population comparable to your patient cohort?
  • Who in your practice is accountable for its outputs?
  • How are clinicians trained to interpret, challenge, and override AI recommendations?

If you can't answer all four, the deployment is premature.

Administrative AI Tools

Administrative AI — covering appointment scheduling, patient communication, billing automation, referral management, documentation assistance, and supply management — sits in a different risk category. The governance requirements are less prescriptive here, but they are not absent.

The key requirements for administrative AI in a health setting are:

  • Patient data handled by the tool must comply with the Privacy Act and your practice privacy policy
  • Any AI tool that communicates directly with patients must disclose its automated nature where relevant
  • Audit trails must be maintained for AI-generated communications or decisions that could affect patient care

This is the area where most independent practices have the most accessible near-term opportunity. Administrative AI automation that meets these requirements can recover substantial clinician and staff time — and unlike clinical AI, the governance pathway is straightforward and well-understood. It's also where the ROI case is clearest: automating appointment reminders, intake processing, and billing workflows delivers measurable time savings without navigating the TGA pathway.

Patient Data and Privacy

The intersection of AI and patient data is the highest-risk area in the framework. NSW Health data governance policies are explicit: patient data should not be transmitted to systems where storage location, access controls, and data handling practices cannot be verified and documented.

This directly affects practices using consumer-grade AI tools — including general-purpose large language models — to process patient records, draft clinical notes, or summarise consultations. Without a formal data processing agreement, documented data flows, and verified Australian data residency or compliant international transfer controls, you are exposed under both the Privacy Act and the framework. The fact that a tool produces useful outputs does not make it compliant.

The Broader Compliance Landscape: TGA, Privacy Act, and Beyond

Understanding how the NSW Health AI Framework fits into the broader regulatory ecosystem matters — particularly in 2026, when that ecosystem has matured considerably compared to even two years ago.

The TGA has refined its SaMD classification guidance through multiple consultation rounds, and enforcement has become meaningfully more active. Practices that have relied on ambiguity in earlier guidance to deploy unregistered clinical AI tools are increasingly at risk of regulatory scrutiny. The it wasn't clear at the time defence has a diminishing shelf life.

The Privacy Act, through reforms introduced following the Privacy Act Review, has strengthened requirements around automated decision-making and the right to explanation — both of which intersect directly with AI use in healthcare. Patients now have stronger grounds to request information about how automated systems have influenced decisions about their care.

The Australian Digital Health Agency's work on FHIR interoperability standards is also relevant. As AI tools increasingly integrate with clinical systems, ensuring those integrations don't create data governance gaps is an active compliance requirement — not a future consideration. The question of which system holds the authoritative record when AI is involved needs to be answered before deployment, not after an incident.

For a broader view of the federal-level picture, our guide to the Australian Government AI Strategy covers the regulatory context that the NSW Health AI Framework sits within — including the emerging mandatory guardrails for high-risk settings that explicitly include healthcare.

What Practices Are Getting Wrong Right Now

Working with healthcare organisations navigating this landscape, the failure modes are consistent — and largely preventable.

No governance documentation. Practices have adopted AI tools — sometimes bundled into existing software updates — without any internal documentation of what the tool does, what data it accesses, who owns it, or how it's monitored. This is a compliance gap, not a minor formality. If you can't produce a one-page governance record for each AI tool in your practice, you aren't compliant with the framework's accountability requirements.

Assuming vendor compliance means practice compliance. A vendor who holds ISO 27001 certification or SOC 2 attestation has not told you their tool is TGA-registered, that it handles data under Australian Privacy Principles, or that it meets NSW Health governance requirements. These are distinct obligations, and they all sit with the practice. Vendor certifications are necessary but not sufficient.

Misclassifying clinical AI as administrative. Some practices have categorised AI tools as administrative to sidestep clinical governance requirements, when those tools are indirectly influencing clinical decisions. If your AI is reading clinical notes to triage patients or prioritise recall lists, it is not purely administrative — and the framework will treat it accordingly. The classification should follow the function, not the preferred governance pathway.

No post-deployment monitoring. The framework explicitly requires ongoing performance monitoring. Most practices that have deployed AI have no formal process for evaluating whether it's continuing to perform as intended after initial rollout. AI tools drift. Vendors push updates. Use patterns change. Without monitoring, you won't know there's a problem until there is one.

A Practical Governance Framework for Your Practice

The framework doesn't demand perfection — it demands documented process. Here's what a compliant, well-managed AI governance posture looks like for an independent practice or clinic in 2026:

Step 1: Conduct an AI inventory. List every AI-enabled tool your practice currently uses, including features embedded in existing software. Many practice management systems now include AI-powered features that weren't there 18 months ago. Categorise each as clinical or administrative and document its data handling practices.

Step 2: Identify governance owners. Every tool on your list needs an identifiable owner — typically a practice principal or practice manager. That person is responsible for monitoring the tool, reviewing incidents, and deciding whether to continue deployment. If nobody owns it, nobody is managing it.

Step 3: Establish a data governance baseline. Review your privacy policy, patient consent forms, and data processing agreements with vendors. Ensure they accurately reflect current AI use and cover all relevant data flows. Gaps in your privacy documentation create exposure under both the Privacy Act and the framework.

Step 4: Address clinical AI compliance first. If any tool could be classified as SaMD, seek TGA advice before proceeding. The regulatory risk here is not theoretical, and proactive compliance costs a fraction of what a compliance investigation will. If you're unsure whether a tool qualifies, that uncertainty is itself a reason to get advice.

Step 5: Build monitoring into operations. Create a simple quarterly review process for each AI tool. Is it still performing as intended? Have there been incidents or near-misses? Has the vendor updated the tool in ways that change its risk profile? Fifteen minutes per quarter per tool is a reasonable commitment for the governance assurance it provides.

Where AI Automation Fits — and What's Safe to Start Now

Here's the genuinely encouraging part: this framework is not a barrier to AI adoption. It's a structure for doing it properly. For most independent practices, the clearest near-term opportunity is administrative AI automation — which sits squarely within the lower-risk category and has a well-defined governance pathway.

Consider what this can include within a compliant framework:

  • Appointment reminders and recalls via SMS or email, freeing reception staff for higher-value patient interactions
  • Patient intake form processing that extracts and routes information before the consultation begins
  • Medicare billing and claim automation that reduces claim errors and accelerates revenue cycle processing
  • Referral letter drafting using clinician-approved templates with mandatory human review before dispatch
  • Clinical supply and inventory management for consumables and medical equipment
  • After-hours patient communication workflows that triage non-urgent queries without clinical involvement

Each of these can be implemented fully within compliance requirements — provided data governance, transparency, and monitoring are built in from day one, not retrofitted later. The governance overhead for administrative automation is manageable. The ROI is real. And the compliance pathway is clear.

Our healthcare supply chain automation case study shows what this looks like in practice, and how quickly practices achieve measurable ROI when automation is scoped and governed correctly from the outset.

For a broader view of what administrative AI automation looks like across healthcare settings, see our guide to AI automation for Australian healthcare providers. And if you're at the stage of thinking about where your practice's AI strategy should go over the next 12 to 24 months, our AI strategy consulting service is designed exactly for that conversation.

Practical Takeaways

  • This governance framework applies to your practice, whether you're a large health district or an independent GP. The obligations scale; the principles don't.
  • Administrative AI automation is your lowest-risk, highest-value starting point. Get the governance right and the ROI follows quickly — typically within the first year.
  • Vendor compliance is not practice compliance. You are accountable for how your tools handle patient data, regardless of your vendor's certifications or contractual indemnities.
  • TGA SaMD obligations are real and actively enforced in 2026. If your AI influences clinical decisions, verify its registration status before your next accreditation cycle does it for you.
  • Documentation is governance. An AI tool without an owner, a monitoring process, and an audit trail is a liability dressed as a solution.
  • Early movers win twice. Practices that build compliant AI governance now will find subsequent adoption faster and cheaper than those retrofitting compliance later — and will be better positioned when accreditation bodies formalise their AI governance requirements.

Ready to Build Compliant AI Automation Into Your Practice?

Navigating the NSW Health AI Framework alongside TGA, Privacy Act, and Australian Digital Health Agency requirements is genuinely complex — particularly for practices without a dedicated compliance function. Getting it right requires both technical capability and regulatory understanding. Getting it wrong means either missing the opportunity entirely or deploying automation that creates more risk than it removes.

Iverel is a Perth-based AI automation agency working with healthcare organisations across Australia to implement AI automation that is commercially valuable and governance-ready from day one. We don't just build the automation — we help you build the documentation, monitoring processes, and accountability frameworks that make your deployment defensible under the framework and beyond.

Whether you're starting with a single administrative workflow or building a comprehensive AI governance programme for your practice, we can help. Explore our process automation services or get in touch to discuss your practice's specific situation.

The practices getting this right in 2026 won't just be compliant — they'll be faster, leaner, and better positioned for whatever the next wave of AI adoption brings.

NSW Health AI Frameworkhealthcare AI governanceTGA SaMDAI compliance healthcareclinical workflow automationAI in healthcare Australiamedical practice AIresponsible AI healthcare

Ready to automate?

Every business is different. Book a call and we'll map out exactly what AI can handle for you.

Book a Free Call →