30 October 2026
New guidance gives Australian Government agencies a shared foundation for designing and governing AI systems. The guidance helps agencies design, procure, and govern AI systems while strengthening trust, accountability, and public value.
The DTA has published the AI Reference Architecture, a foundation for agencies designing and implementing artificial intelligence across the Australian Government.
This resource translates government policy and technical expectations into reusable architectural structures. Agencies can adapt these structures to their objectives and operating environments, while also being able to apply them according to their risk profiles.
As agencies adopt AI, technical decisions about data, models, integrations, vendors and human oversight can have significant implications for several factors impacting the long-term public good, such as: privacy, accountability, fairness, safety, and interoperability.
Yet the consequences of those decisions are often not visible until after commitments have been made. This reference architecture helps agencies make the trade-offs more visible and understandable before committing to a solution, enabling more informed discussions.
AIRA helps agencies design trustworthiness and assurance into AI systems from the beginning, as opposed to retrofitting them after key technology decisions have already been made.
“Artificial intelligence presents enormous opportunities to improve services and outcomes for Australians,” explains Lucy Poole, Deputy CEO - Digital Strategy, Planning and Performance. “But agencies need practical tools to make informed decisions about how these technologies are adopted and governed. The challenge for agencies is designing systems that meet their needs while the technology keeps evolving.”
“Designing an AI system has a lot in common with designing a building. Every structure needs certain fundamentals: a solid foundation, core utilities, clear points of access, and safety overrides if things go wrong. And those elements will look very different depending on whether you are designing a cottage, a skyscraper, or a warehouse.”
“The AI Reference Architecture works in a similar way. It gives agencies a common foundation for AI without prescribing the final design. Agencies can use that foundation to design systems that reflect their own requirements, services and operating environments.”
That common foundation brings together 4 architectural constructs.
The functional stack explains the main building blocks needed to operate an AI system across a set of 5 interconnected layers.
The experience layer covers how people, other systems, and devices all interact with AI. The application and orchestration layer coordinates workflows, business rules, AI models, human decisions and external services.
The intelligence and analytics layer produces recommendations and insights. It covers how AI models are developed, tested, deployed, monitored and eventually retired.
The data and metadata layer manages the information these models use. This includes collecting, storing, preparing, sharing and tracking data, even when its quality is imperfect.
Finally, the platform and infrastructure layer provides the computing power, storage, networks and hosting environments needed to run everything securely.
AI systems need to connect with people and other technology to work. The guidance helps agencies make those connections without exposing sensitive information or losing control of how data is used.
“Robust information sharing is essential to building AI systems that work across government,” stresses Ms Poole. “It must protect sensitive information and maintain clear trust boundaries, without creating barriers to integration.”
“This means supporting secure connections across legacy and modern systems, while remaining flexible enough to work with diverse internet-connected devices, sensors and services.”
Our reference architecture outlines how agencies can build confidence in AI systems from the start. Rather than checking for problems only after deployment, agencies should design safeguards into every part of the system.
This means continuously monitoring how the AI performs and keeping records of its data, models, and decisions. Agencies should be able to explain how the system works and what its outputs mean. The AI must also remain reliable, fit for purpose and safe when conditions change or errors occur.
The reference architecture also calls for agencies to identify and reduce bias. Services should remain accessible for people with different needs and circumstances. Security, privacy and resilience must apply throughout the AI system’s lifecycle.
Agencies should consider long-term sustainability. This includes environmental impacts, operating costs and dependence on external providers.
The guidance calls for clearly assigned roles and suitable skills. Higher-risk or irreversible actions may require additional review or escalation. When AI helps draft or analyse material, people should still verify the output and retain responsibility for the final work.
People must be able to monitor, pause, change or stop systems within appropriate timeframes. Pre-approved safeguards can also act automatically, such as pausing operations or reversing changes when risks emerge.
“Systems should be designed to reflect how people actually interact with automation. This includes managing risks such as over-reliance on AI, reduced attention during prolonged monitoring and the gradual loss of human expertise,” states Ms Poole.
The AI Reference Architecture is supported by a companion guide focusing on scenarios that show how agencies can apply the architecture in practice. Each scenario helps agencies identify where control sits. The scenarios also highlight where further evidence or safeguards may be needed.
Human oversight remains a central consideration in each scenario. The examples show how it can be designed into an AI-enabled system.
“The AI Reference Architecture brings together disparate pieces together in the application and roll out of AI service,” concludes Ms Poole. “It explains the main parts of a system, what each does and how they connect. Agencies can use it as a starting point for their own solution architecture, adapting it to their needs, existing technology and risks.”
“It brings relevant government requirements and technical expectations into one place, providing technical guardrails for system design. These help agencies plan connections, safeguards and human oversight before committing to particular technologies.”
“Using the architecture is voluntary. It supports agencies to meet applicable requirements, but does not replace them.”
To view the full AI Reference Architecture and supporting AI Reference Architecture Companion Guide - Scenarios, visit digital.gov.au.
The Digital Transformation Agency is the Australian Government's adviser for the development, delivery, and monitoring of whole-of-government strategies, policies, and standards for digital and ICT investments, including ICT procurement.
For media enquiries email us at media@dta.gov.au
For other enquiries email us at info@dta.gov.au