Agile InsightsBriefing
A closed reference volume, a folded paper map with a brass compass and a pair of headphones on a dark desk, labelled The Map, Reference Series and Reference 0. A yellow bar along the bottom reads Reference series, number 0.

Reference 0 · Primer · About 7 minutes

The AI governance stack: a map of the reference series

How the standards, frameworks and rules fit together in three layers, and where to start for your role.

#rules

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

In this episode

01Four kinds of document, not oneSome are laws you must follow, some are standards you can be certified against, some are voluntary guidance and some are bodies of knowledge. Knowing which kind you are looking at is half of the answer.
02Three layers, pictured as a poolData foundations are the water, AI governance is the swimmer, and law and regulation are the pool rules. For government, the DTA's AI Reference Architecture is the lane ropes between the rules and each agency's design.
03The question each layer answersThe rules layer asks what you must do. The governance layer asks how you run AI so you can show you did it. The data layer asks whether you can trust what goes in.
04How the series stays currentMost episodes were written in October 2026 and the government episodes in November, from public sources, with a review planned each quarter. Standards change slowly, while the rules and the agent guidance change fast.
05Where to start for your roleBoards start with the rules, risk and security leaders with ISO/IEC 42001 and NIST, data owners with the two data episodes. Government and its suppliers start with References 13, 12 and 14, then 15.

Written from public sources, checked on 10 October 2026 and updated on 11 November 2026. The standards are paraphrased, not reproduced. Check the current text before relying on it.

Our fact check · 11 claims: 9 confirmed

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 Privacy Act duty on automated decisions starts on 10 December 2026Office of the Australian Information Commissioner, transparency resources
  2. ConfirmedAustralia's national AI guidance sets out six essential practicesDepartment of Industry, Science and Resources, Guidance for AI Adoption (October 2025)
  3. ConfirmedISO/IEC 42001 is a management system standard an organisation can be certified againstISO; see Reference 3
  4. ConfirmedThe NIST AI Risk Management Framework gives a shared language for AI riskNIST; see Reference 4
  5. ConfirmedISO/IEC 23894 is guidance on managing AI riskISO; see Reference 7
  6. ConfirmedThe Digital Transformation Agency published the AI Reference Architecture (AIRA) in October 2026DTA, AIRA v1.0; see Reference 12
  7. ConfirmedAIRA mandates no products or designs, but agency architectures are expected to be in line with itAIRA v1.0, p. 20
  8. ConfirmedThe Commonwealth AI rulebook is the responsible use policy, the AI technical standard and the AI Plan for the APSDTA policy pages; see Reference 13
  9. ConfirmedSecurity for government AI rests on the Protective Security Policy Framework and the Information Security ManualHome Affairs and the Australian Signals Directorate; see Reference 14
  10. Data as the water, governance as the swimmer, rules as the pool and AIRA as the lane ropes is a teaching pictureOur analysis; —
  11. Standards change slowly; the rules and the agent guidance change fastOur analysis; —

How this maps to what we do

Three layers, from the data underneath, through the governance of AI itself, to the rules that apply. For government, a fourth piece now sits between the rules and each agency's own design: the DTA's AI Reference Architecture. Each episode in this series covers one piece.

Standard or rule
What an organisation needs
Where we help
Data you can trust for the use cases you plan.
Data Foundation ReadinessJust-in-Time Data Governance
Policy to solution (government): the DTA AI Reference Architecture (AIRA)Episode: The DTA AI Reference Architecture: one blueprint for government AI
One shared design that turns the rules into AI systems an agency can build, buy and assess.
AI DiagnosticAI Governance Framework Alignment
Starting point
A short baseline of where you stand before choosing where to build.
AI DiagnosticAgile Yarn

These are Agile Insights AI Strategy & Governance offerings.

Want to know more?

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

Read the transcript

Leonard There are dozens of standards, laws and frameworks that claim to govern AI, and nobody has time to read them all. This episode is a map. By the end you will know how they fit together in three layers, which ones matter first for your role, and where to start listening. This is as at November 2026.

Leonard First, what the series is. It is a set of short reference episodes from Agile Insights, each about five or six minutes, each covering one standard, framework or rule. Every episode follows the same pattern: what it is, the problem it solves, who uses it, what is in it, how to get started, and where it stands today. Each one is checked against the publisher's or regulator's own pages before it goes out, and each says plainly that the current text must be checked before you rely on it.

Leonard The problem the map solves is a familiar one. A board asks, are we compliant with AI regulation? An executive hears four names in one meeting, a standard, a framework, a law and a guidance note, and they all sound like the same thing. They are not. Some are laws you must follow. Some are standards you can be certified against. Some are voluntary guidance. Some are bodies of knowledge. Knowing which kind you are looking at is half of the answer.

Leonard Who is it for? Directors, chief executives, and the risk, privacy, security, data and technology leaders who brief them. It is written for people who need to make decisions and ask good questions, not for people who will implement every clause.

Leonard Now the map itself, in three layers. Think of a swimming pool. The bottom layer is the water. That is your data foundations. AI is only as trustworthy as the data under it. Two episodes cover this layer: the data management frameworks D A M A D M B O K and D C A M, and the I S O, I E C, 5259 series on data quality for machine learning.

Leonard The middle layer is the swimmer. That is AI governance itself: how an organisation owns, runs, checks and improves its AI. Six episodes sit here. I S O, I E C, 42001 is the management system standard you can be certified against. The N I S T AI Risk Management Framework gives you a shared language for risk. I S O, I E C, 23894 is guidance on how to manage AI risk. I S O 27001 and 27701 are the security and privacy standards that many organisations already hold, and that give you a head start. Then two episodes on agents: governing AI agents in general, and the Microsoft Agent Governance Toolkit, a tool that enforces rules at runtime.

Leonard The top layer is the pool rules. That is the law and regulation that applies where you operate. Here you will find the Privacy Act duty on automated decisions that starts on the tenth of December 2026, Australia's national AI guidance and its six essential practices, the European Union AI Act, the sector rules for industries such as government, finance and health, and work health and safety duties for digital work systems. For the Commonwealth, two more episodes sit here. One covers its own AI rulebook: the responsible use policy, the AI technical standard and the plan for the public service. The other covers security for government AI: the protective security framework and the Information Security Manual. Rules are the layer that can change quickly, so check these episodes first for currency.

Leonard Between the pool rules and the swimmer, government now has a fourth piece: the lane ropes. That is the D T A's AI Reference Architecture, called AIRA, published in October 2026. It turns the rules above into a shared design that every agency's AI system can follow. It mandates no products or designs. But agency architectures are expected to be in line with it. Reference 12 covers it.

Leonard There is one more episode that sits outside the layers: a balanced look at the environmental footprint of AI, in the briefing series under the title, are data centres evil.

Leonard How does this connect to AI in your organisation? Each layer answers a different question. The rules layer asks, what must we do? The governance layer asks, how do we run AI so that we can show we did it? The data layer asks, can we trust what goes in? An organisation that only studies the rules ends up with paperwork. One that only builds governance may miss a legal deadline. One that ignores the data finds out the hard way.

Leonard Where should you start? Four paths. If you are on a board or in the executive team, start with the rules: the Privacy Act episode, then Australia's national guidance, then the European Union AI Act if you have European exposure. If you run risk, security or privacy, start with the governance layer: 42001, then the NIST framework, then the agent episodes. If you own data, start at the bottom with the two data episodes, then listen to 42001 to see where your work plugs in. If you work in government, or sell to it, take the fourth path. Start with Reference 13, the Commonwealth rulebook. Then Reference 12 on AIRA, then Reference 14 on security. Reference 15 is coming next, on the quieter duties: records, data ethics, digital experience and cloud. Wherever you start, the show notes for each episode link to the primary sources.

Leonard A word on currency, as at November 2026. Most of these episodes were written in October 2026 from public sources. The government episodes followed in November. The standards themselves change slowly. The rules and the agent guidance change fast. We plan to review the series each quarter, and the show notes for each episode carry the date it was checked. Always check the current text before relying on anything you hear.

Leonard To recap: data is the water, governance is the swimmer, and the rules are the pool. For government, AIRA is the lane ropes. If you'd like to know more, sign up for our podcast at agile-insights.com.au.

More on #rules

Reference 15 · PrimerThe quiet obligations: records, data ethics, digital experience and cloudReference 14 · PrimerSecurity for government AI: the PSPF, the ISM and the frontier AI eraReference 13 · PrimerThe Commonwealth AI rulebook: the Policy, the technical standard and the APS AI PlanReference 12 · PrimerThe DTA AI Reference Architecture: one blueprint for government AIEpisode 21 · DiscussionTrust cannot be rushed: how much authority will you give AI?Reference 11 · PrimerAustralian sector rules for AI: what applies to your industry

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