Agile InsightsBriefing
A closed reference volume, a partly unrolled blueprint held flat by a steel weight and a pair of headphones on a dark desk, labelled DTA AIRA, Reference Series and Reference 12. A yellow bar along the bottom reads Reference series, number 12.

Reference 12 · Primer · About 6 minutes

The DTA AI Reference Architecture: one blueprint for government AI

What the Commonwealth's new AI Reference Architecture asks of agencies and suppliers, and where it stops.

#rules#oversight

Download the MP3 · Linda, an AI voice, reads a script written from the sources below.

In this episode

01What AIRA is, and why it is neither law nor optional.
02The fragmentation problem it solves.
03Who uses it, and why the 15 December deadline matters.
04Five layers, nine trust domains, human oversight and control points.
05How to get started, and where it stands as at November 2026.
Sources and links (8)
AI Reference Architecture · digital.gov.au
The DTA's home page for AIRA v1.0.
Architecture overview · digital.gov.au
The foundational view: stack, communication, trust domains and human oversight.
Trustworthiness and assurance · digital.gov.au
The nine domains, each defined from a named standard.
Human agency, oversight and control · digital.gov.au
Why oversight is not a human in every loop.
Companion Guide: Scenarios · digital.gov.au
Four worked examples, from Copilot to machine learning in case management.
Relationship with other frameworks · digital.gov.au
Appendix B: the policies AIRA translates, and where it sits beside TOGAF and Zachman.
Policy for the responsible use of AI in government · digital.gov.au
The policy AIRA exists to implement, with the 15 December 2026 deadline.
Industry view: "the right call" · Cyber Daily, 6 Nov 2026
Kinetic IT's chief technology officer on governing AI as one service.

Written from the DTA's published documents and checked on 11 November 2026. AIRA is paraphrased, not reproduced (CC BY 4.0). Check the current text before relying on it.

Our fact check · 20 claims: 19 confirmed, 1 reported

Before we record, every claim the hosts make is checked against the sources (this table was last updated 11 November 2026). Confirmed means a primary source or two independent reports; Reported means one party says so; Partly right, Disputed, Unverified and Not yet known mean what they say, and the hosts say so on air. This is the table as we left it.

  1. ConfirmedThe Digital Transformation Agency published the AI Reference Architecture (AIRA), version 1.0, in October 2026, with a companion guide of four worked scenariosDTA media release and digital.gov.au AIRA pages; Companion Guide: Scenarios v1.0
  2. ConfirmedAIRA mandates no products, platforms or designsAIRA v1.0, Introduction
  3. ConfirmedAgency architectures "should be in line with" AIRA, so it is expected in structure and language rather than voluntaryAIRA v1.0, p. 20
  4. ConfirmedAIRA answers fragmentation: plentiful policy, law and standards that each agency had to read and reconcile aloneAIRA v1.0, Appendix B
  5. ConfirmedThe obligations it translates include the responsible use of AI policy, the Privacy Act, the Archives Act, the Protective Security Policy Framework and the Information Security ManualAIRA v1.0, Appendix B
  6. ConfirmedOne use is like-for-like comparison of vendor offerings in procurementAIRA v1.0, Introduction (ways to use AIRA)
  7. ConfirmedIts users include executives, architects, procurement teams and risk, audit and cyber teams; governance teams use it as a common baseline for approval or an authority to operateAIRA v1.0, Appendix A
  8. ConfirmedBy 15 December 2026 Commonwealth agencies need a register of AI use cases with named owners, staff training and impact assessments under wayDTA Policy for the responsible use of AI in government v2.0
  9. ConfirmedThe functional stack has five layers: experience; application and orchestration; intelligence and analytics; data and metadata; platform and infrastructureAIRA v1.0, Architecture description
  10. ConfirmedThere are nine trustworthiness and assurance domains, including fairness, privacy, safety, traceability and sustainabilityAIRA v1.0, Trustworthiness and assurance (an overview figure lists eight; the domain page and the second figure list nine)
  11. ConfirmedSustainability covers dependence on vendors and the ability to replace key technologies over timeAIRA v1.0, Trustworthiness and assurance, Sustainability
  12. ConfirmedHuman oversight does not mean a person approving every action; it means automation within limits people can watch, pause, revise and revokeAIRA v1.0, Human agency, oversight and control
  13. ConfirmedEach part of a system is a control point the agency governs, configures or designsCompanion Guide v1.0, control points
  14. ConfirmedFor Microsoft 365 Copilot, more control points are govern only than anything else (23 govern, 16 configure, 8 not exercised)Companion Guide v1.0, Figure 1, p. 9 (counted by us)
  15. ConfirmedAIRA sits between policy and solution design, tracing the requirements of the Ethics Principles, the responsible use policy and the AI technical standard into architectureAIRA v1.0, Appendix B
  16. ConfirmedWhere guidance already exists, such as for security and privacy, AIRA defers to itAIRA v1.0, p. 15 (naming the Information Security Manual and the privacy regulator's advice as examples is our reading)
  17. ConfirmedAIRA never mentions ISO/IEC 42001No mention in either PDF (searched 11 Nov 2026)
  18. ConfirmedUsers who cannot or will not use AI should keep a non-AI way to the serviceAIRA v1.0, Architecture description, non-AI alternatives
  19. ConfirmedCapability groups and components are expected to evolve as technology, standards and practice matureAIRA v1.0, Architecture overview
  20. ReportedIndustry commentators expect AIRA's language in government tendersAivy.com.au, 3 Nov 2026

How this maps to what we do

The rules that apply to AI in Australia and for Australian organisations with European exposure, and the practical work each one calls for.

Standard or rule
What an organisation needs
Where we help
Privacy Act: automated decision-making transparency (from 10 December 2026)Episode: Is your AI making decisions about people? What the Privacy Act now asks of you
A list of the automated decisions made about people, including those buried in vendor software, and the policy wording that describes them.
AI Risk & ComplianceUse Case Portfolio & AssessmentAI Governance Framework Alignment
Australia's AI guidance: the six essential practices (with VAISS, SafeAI-Aus and NAIC)Episode: Australia's AI guidance: the six essential practices
A baseline against the six practices, then a prioritised roadmap to close the gaps.
AI DiagnosticAI Governance Framework AlignmentAI Strategy Build
Each AI system classified by risk tier, with the obligations and dates that apply to it, including for non-EU organisations.
AI Risk & ComplianceUse Case Portfolio & AssessmentAI Governance Framework Alignment
Australian sector overlays: government, health, finance, education, disability, aged care, not-for-profitEpisode: Australian sector rules for AI: what applies to your industry
The sector's own obligations mapped onto one set of AI controls, so that nothing is done twice.
AI Risk & ComplianceAI Governance Framework Alignment
A review of worker monitoring and automated decisions about people for safety and psychosocial risk.
AI Risk & ComplianceAdoption & Engagement

These are Agile Insights AI Strategy & Governance offerings. Which of them you need depends on where your organisation starts.

Want to know more?

Sign up for our podcast at agile-insights.com.au.

Read the transcript

Linda The federal government now has one blueprint for how every AI system in the Commonwealth should be designed, bought and governed. It is called the AI Reference Architecture. By the end of this episode, you will know what it asks of agencies. You will also know what it means for suppliers, and where it stops.

Linda First, what it is. The Digital Transformation Agency, the D T A, published the AI Reference Architecture in October 2026. People call it AIRA. It is a reference architecture, which means a shared structure and vocabulary, not a product list. It mandates no products, platforms or designs. But the agency says agency architectures "should be in line with" it. So it is not a law, and it is not optional in the casual sense either. It is the expected shape. A companion guide with four worked scenarios comes with it.

Linda The problem it solves is fragmentation. A department running a chatbot must satisfy the responsible use of AI policy, the Privacy Act and the Archives Act. Then come the protective security framework and the Information Security Manual. Until now, each agency read those rules on its own and drew its own conclusions. AIRA translates them into one set of reusable patterns. Picture two agencies buying the same call centre product. Without a common frame, each writes its own assessment from scratch. With one, they can compare like with like.

Linda Who uses it? Chief AI officers and executives use it to frame strategy and investment. Architects use it to design and review systems. Procurement teams use it to compare vendors. Risk, audit and cyber teams use it as a common baseline. That matters when an architecture board meets, or a system needs its authority to operate. The timing matters too. By the fifteenth of December 2026, Commonwealth agencies need a register of their AI use cases and named owners. They also need staff training, and impact assessments under way. AIRA gives them a way to describe the systems they are assessing.

Linda So what is in it? Four ideas carry it. The first is a functional stack of five layers. They are experience, application and orchestration, intelligence and analytics, data and metadata, and platform and infrastructure. Every AI system touches all five. The second is nine trustworthiness and assurance domains. They include fairness, privacy, safety, traceability and sustainability, and they are designed in, not bolted on afterwards. Sustainability here includes how dependent you are on one vendor, and how you would leave.

Linda The third idea is human agency, oversight and control. AIRA is clear that this does not mean a person approving every action. It means automation running inside limits that people can watch, pause, revise and revoke in time. The fourth idea is control points. For each part of a system, an agency either only governs it, configures it, or designs it. For a product like Microsoft 365 Copilot, more of the system is govern only than anything else. So the real work is evidence, contract terms and the few settings you actually control.

Linda How does it connect to the rest of the stack? AIRA sits between policy and solution design. Above it sit the AI Ethics Principles, the responsible use policy and the AI technical standard, whose requirements AIRA traces into architecture. Below it sits each agency's own solution. For security and privacy, it says it will defer to existing guidance, such as the Information Security Manual and the privacy regulator's advice. It never mentions I S O, I E C, 42001, the AI management system standard. So any link to that standard is one you draw yourself.

Linda Getting started. First, list your AI use cases and draw each one on the five layers. Second, for each part, mark whether you govern, configure or design it, and be honest about how much is govern only. Third, check that you can roll back models, data and settings. And make sure people who cannot or will not use AI still have another way in. The primary sources are the AI Reference Architecture and its companion guide on the digital dot gov dot A U website. Both are linked in the show notes.

Linda Now, currency. As at November 2026, AIRA is at version one point zero, published in October. The D T A says its capability groups and components are expected to evolve, and agency experience feeds back into it. Expect revisions, and more detailed patterns. Check the current text before you rely on any detail here.

Linda In short, AIRA is the shape every Commonwealth AI system is now expected to take. Industry commentators expect its language in government tenders. If you'd like to know more, sign up for our podcast at agile-insights.com.au.

More on #rules

Reference 13 · PrimerThe Commonwealth AI rulebook: the Policy, the technical standard and the APS AI PlanReference 15 · PrimerThe quiet obligations: records, data ethics, digital experience and cloudReference 14 · PrimerSecurity for government AI: the PSPF, the ISM and the frontier AI eraEpisode 21 · DiscussionTrust cannot be rushed: how much authority will you give AI?Reference 11 · PrimerAustralian sector rules for AI: what applies to your industryReference 8 · PrimerISO/IEC 27001 and 27701: security and privacy, for AI too

Every episode tagged #rules

All of the reference series (16) · Follow AI Governance Reference in Apple Podcasts or any podcast app

Everything from Agile Insights: briefings, Making You Think and the reference series