ISO 42001 in Australia: A Practical Implementation Guide

ISO/IEC 42001 in Australia: the steps, a realistic timeline, and how it fits the Privacy Act and the Voluntary AI Safety Standard.

ISO/IEC 42001 is the international standard for an AI management system (AIMS). In Australia it is published as AS ISO/IEC 42001:2023. Most pages that rank for it explain what the standard is. This guide is for the next question: you have decided it is relevant, so what does it take to actually implement it in an Australian organisation, in what order, and how does the work line up with the Privacy Act and the government's AI guidance?

It is deliberately different from two guides we already publish. AI Governance Australia is the board-level view: policies, risk tiers and oversight habits that let leaders set guardrails. AI, Auditors, and ISO Certification explains how compliance automation, qualified auditors and accredited certification bodies fit together. This article sits between them and is only about standing up the management system that ISO/IEC 42001 describes: scope, steps, evidence, timeline and how it connects to Australian expectations.

We work from Burleigh Heads on Queensland's Gold Coast, remotely, with no city offices, and we have been building software and AI systems for Australian businesses since 2016. This guide is general information rather than legal, privacy or audit advice. Everything we state about the standard and about Australian guidance comes from official pages we read on 31 October 2026, listed under Sources at the end.

What ISO/IEC 42001 actually is

ISO describes ISO/IEC 42001 as the international standard that specifies requirements for establishing, implementing, maintaining and continually improving an Artificial Intelligence Management System within an organisation. It is written for organisations that provide or use AI-based products or services, and ISO calls it the world's first AI management system standard.

The word that matters is management system. ISO is explicit that this is not a standard about the details of a particular AI application. It is a management system standard: you put policies, objectives and processes in place for the sound governance of AI across the organisation, run on the Plan-Do-Check-Act cycle. ISO's own wording is that it gives a practical way of managing AI-related risks and opportunities across an organisation rather than examining individual applications one by one.

Standards Australia adopted the standard as AS ISO/IEC 42001:2023 and announced it on 16 February 2024. Its announcement describes the standard as outlining internal governance and risk management, and says it can be used by any size or type of organisation that intends to use AI to develop or deliver products and services.

Two things the standard is not:

It is not a software tool or a checklist of technical settings. It tells you what a managed approach to AI must cover; you decide how your organisation meets it.
It is not certification by itself. ISO states that it does not provide certification or conformity assessment. If you want a certificate, an independent external certification body has to assess your system.

Who needs it, and who does not

ISO's FAQ says the standard is for organisations of any size that are involved in developing, providing or using AI-based products or services, across all industries, including public sector agencies, companies and non-profits. That is broad on purpose. In practice, the organisations that get the most from implementing it tend to fall into a few groups:

Organisations that build or supply AI to other businesses, where customers and procurement teams increasingly ask how AI risk is managed.
Organisations that use AI in decisions about people, such as hiring, customer eligibility, pricing, triage or fraud screening, where the consequences of a bad outcome land on an individual.
Organisations already running other management systems (quality, information security, safety) that want AI handled with the same discipline and audit trail.
Organisations that must show a board, a regulator or a major customer that AI use is governed rather than improvised.

Equally honest: not every organisation needs to certify. None of the official pages we read describe ISO/IEC 42001 certification as a legal requirement for Australian organisations, and the Voluntary AI Safety Standard is, as its name says, voluntary. A small team using a handful of low-risk AI tools may be better served by a lighter governance baseline first (see AI Governance Australia). Check your own contracts, tender conditions and sector regulators, because a customer or regulator can still ask for specific assurance.

The implementation path, step by step

This is the sequence we would follow for an Australian organisation starting from scratch. It is a practical path, not the clause text of the standard, so buy and read the standard itself before you build against it.

1. Decide why you are doing it, and what is in scope

Start with the reason. Is it a customer requirement, a board request, a risk concern, or a desire to be ready for tighter rules? The reason shapes the scope. Define which parts of the organisation, which AI systems and which roles (you may be a developer, a provider, a user, or all three) the management system will cover. A narrow, honest scope that you can evidence beats a broad one you cannot.

2. Get leadership commitment and name accountability

A management system lives or dies on ownership. Name an accountable executive, decide who runs the system day to day, and set the organisation's AI policy and objectives. This is where the board-level material in AI Governance Australia feeds straight in: the policy, the oversight rhythm and the reporting pack are inputs, not duplicates.

3. Build an inventory of AI systems

You cannot manage what you cannot list. Create a register of every AI system in scope: what it does, who owns it, which data it touches, which supplier or model sits behind it, and who is affected by its outputs. Include the unofficial ones, such as browser extensions and personal accounts used for work. The Australian Government's Guidance for AI Adoption includes an AI register template for exactly this reason (see below).

4. Assess risks and impacts

For each system, work out what can go wrong and for whom: inaccurate outputs, bias, privacy exposure, security weaknesses, misuse, over-reliance, supplier failure. ISO describes the standard as covering the full path from risk assessment through to effective treatment of those risks. Record the method you use, rate the risks consistently and decide what you will do about each one. Treat the people affected by a system as part of the assessment, not an afterthought.

5. Choose and implement controls, and record why

Select the controls that address your risks, implement them, and keep a record of which controls apply to your scope and why. Typical areas include data quality and provenance, human oversight, testing and evaluation before release, monitoring after release, supplier and third-party diligence, transparency to users, and incident handling. This is the stage where technical delivery matters most: logging, evaluation sets, review queues and access controls have to exist in the systems, not only in a document.

6. Operate the system and gather evidence

A policy only counts if people follow it. Train the people who build, buy and use AI, run the processes you defined, and keep the records that show it happened: approvals, test results, incident logs, supplier checks, change records. Evidence is a by-product of normal work if you design the work well. If you have to reconstruct it before an audit, the process is not working.

7. Audit yourself and review as a leadership team

Before anyone external looks, run an internal audit against your own system and the standard, raise the gaps and close them. Then hold a management review where leadership looks at performance, incidents, risks, audit findings and whether the system still fits the business. The Plan-Do-Check-Act cycle only works if the Check step is real.

8. Decide on certification, then maintain

If you want independent assurance, engage an external certification body. Certification bodies typically assess in stages and then return at intervals, so treat it as an ongoing cycle rather than a one-off event. Confirm the body's accreditation status and its scope for this standard directly with them before you commit. Whether or not you certify, keep improving: new systems, new suppliers and new rules all change your risk picture.

A realistic timeline (without made-up numbers)

Anyone who quotes you a fixed number of weeks without looking at your organisation is guessing. Duration depends on a handful of things you can assess yourself, so here is how to think about it instead.

The work moves through the same phases in every organisation, and each phase has to finish enough for the next to be meaningful:

Scoping and ownership. Short if leadership is aligned, long if nobody wants to own AI.
Inventory, risk and impact assessment. Scales with the number and complexity of AI systems and with how many teams use AI informally.
Controls and documentation. Scales with how much already exists. Organisations with mature quality or security management systems usually have a head start on document control, internal audit and management review.
Operating period. You need a genuine period of running the system so there is real evidence. It cannot be compressed by writing more documents.
Internal audit and management review. Must happen before an external assessment, and findings must be closed.
External assessment, if you choose it. Scheduled around the certification body's availability, not only your readiness.

What typically makes it longer: many AI systems, heavy dependence on third-party models and suppliers, no existing management system, unclear ownership, and AI used in decisions that affect individuals. What typically makes it shorter: a narrow initial scope, an existing certified management system, an up-to-date AI register and a named owner with authority.

A sensible way to de-risk the timeline is to start with a small, well-defined scope, run it properly, and extend it as the system matures, rather than attempting every system at once.

How ISO/IEC 42001 fits with the Privacy Act

The Privacy Act 1988 and the Australian Privacy Principles apply whether or not you hold any AI standard. ISO/IEC 42001 does not replace them, and implementing it does not by itself make you compliant. What the management system gives you is the structure to meet privacy obligations consistently when AI is involved.

The Office of the Australian Information Commissioner (OAIC) is clear about the baseline in its guidance on commercially available AI products. Its key points include:

Privacy obligations apply to any personal information put into an AI system, and to the output where it contains personal information.
When adopting a commercially available product, organisations should do due diligence on whether it suits the intended use, including whether it has been tested for that use, how human oversight can be built in, the privacy and security risks, and who will have access to the information.
Privacy policies and notifications should be clear about the use of AI, and public-facing tools such as chatbots should be clearly identified as AI.
If AI generates or infers personal information, that is a collection of personal information that must comply with APP 3. Inferred, incorrect or artificially generated information about an identifiable person, including hallucinations and deepfakes, is personal information.
APP 6 limits using or disclosing personal information put into an AI system to the primary purpose it was collected for, unless consent or a reasonably expected related secondary use applies.
As best practice, the OAIC recommends not entering personal information, particularly sensitive information, into publicly available generative AI tools.

There is also a date to plan around. The OAIC says that from 10 December 2026, APP entities must include information in their privacy policy where they have arranged for a computer program to make, or do something substantially and directly related to making, a decision that could reasonably be expected to significantly affect an individual's rights or interests and that uses the individual's personal information. The policy must cover the kinds of personal information used and the kinds of decisions made solely, or substantially informed, by the program. The OAIC has published a fact sheet, a flowchart and updated APP 1 guidelines to help.

This is where an AI management system earns its keep. The register from step 3, the impact assessments from step 4 and the oversight records from step 6 are exactly what you need to answer "which of our systems make or inform significant decisions about people, using which personal information?" Without that inventory, writing an accurate privacy policy is guesswork. Treat the management system as the engine and the privacy policy as one of the things it produces. Get legal advice on whether and how the new obligation applies to you.

How it fits with Australia's AI guardrails

The Australian Government's Voluntary AI Safety Standard (VAISS) gives practical guidance to organisations on using and developing AI safely and responsibly. It consists of 10 voluntary guardrails that apply across the AI supply chain. Two points matter here:

The standard's own introduction says it draws on and is aligned with international standards, and names AS ISO/IEC 42001:2023 as the most important of them, alongside the US NIST AI Risk Management Framework. Each requirement in the guardrails is aligned with relevant international and local standards or practices.
On 21 October 2025 the National AI Centre published the Guidance for AI Adoption, which sets out 6 essential practices and, in the Department's words, evolves the Voluntary AI Safety Standard. business.gov.au describes it as having Foundations, for businesses new to AI, and Implementation Practices, for more detailed governance and oversight, with templates including an AI policy template and an AI register template.

The guardrails are voluntary. They also set expectations: the standard's introduction says it sets expectations for what future legislation may look like. So implementing a management system now is one way to be ready, whichever direction regulation takes.

Below is our own practical mapping from the 10 guardrails (as worded on the Department's page) to what an AI management system produces. It is not an official crosswalk, and it is not proof of conformity with either document.

VAISS guardrail (summary)	What your AI management system gives you
1. Accountability process, governance, capability, regulatory strategy	Leadership commitment, named owners, AI policy, competence and training records
2. Risk management process	Documented risk method, risk register and treatment decisions
3. Protect AI systems; data governance, quality and provenance	Data controls, security measures and supplier data checks tied to each system
4. Test models and systems; monitor after deployment	Pre-release testing criteria, monitoring and performance review records
5. Human control or intervention	Defined oversight points, escalation paths and override records
6. Inform end users about AI decisions, interactions and content	User disclosures and labelling decisions, linked to your privacy notices
7. Processes for people affected to challenge use or outcomes	A complaints and challenge route with logged outcomes
8. Transparency across the AI supply chain	Supplier assessments, contract terms and information you give downstream
9. Keep records so third parties can assess compliance	Document control, internal audit and retained evidence
10. Stakeholder engagement, with a focus on safety, diversity, inclusion and fairness	Interested-party analysis and impact assessments that include fairness

Used together, the pattern is simple. VAISS and the Guidance for AI Adoption tell you what the Australian Government expects good practice to look like. ISO/IEC 42001 gives you a management system that runs, audits and improves that practice, and can be independently assessed. The privacy rules are the legal floor under both.

Where Adaptive Media helps

We are not a certification body, and we do not certify anyone's system. Certification needs an independent external body, and independence is the point. What we do is the practical work around it, delivered remotely from Burleigh Heads:

Scoping and gap review. Working out which AI systems, suppliers and teams belong in scope, and how far your current practice is from the standard.
AI inventory and risk workshops. Building the register and running risk and impact assessments with the people who actually use the systems.
Policy and control design. Drafting an AI policy, oversight rules and supplier questions that people can follow, and mapping them to the Australian guidance and your privacy obligations.
Technical controls. Because we build AI systems, we can put logging, evaluation sets, human review steps and access controls into the systems themselves, so the evidence is produced by normal operation.
Readiness for an internal audit. Preparing the records and running a mock review before an external party looks.

If senior technology ownership is the missing piece, a retainer such as CTO as a Service can sit alongside this work. If you are not yet sure your organisation is ready for AI at all, start with an AI readiness assessment, and if the goal is moving a pilot into production under these controls, see AI implementation in Australia.

Frequently asked questions

What is ISO/IEC 42001?

It is the international standard that specifies requirements for establishing, implementing, maintaining and continually improving an AI management system within an organisation. ISO describes it as the world's first AI management system standard. In Australia it is published as AS ISO/IEC 42001:2023.

Is ISO 42001 mandatory in Australia?

None of the official pages we read describe it as a legal requirement for Australian organisations, and the Voluntary AI Safety Standard is, as its name says, voluntary. That said, a customer, tender or regulator may ask for it or for equivalent assurance, and the Privacy Act applies to AI use regardless. Check your own obligations and take legal advice where it matters.

What is the difference between ISO 27001 and ISO 42001?

Both are management system standards, but they cover different subjects. ISO/IEC 27001 is the information security management standard; ISO/IEC 42001 is about governing the development, provision and use of AI. Organisations that already run one management system often find the mechanics of the other familiar, but one does not replace the other.

Is there ISO 42001 certification available in Australia?

ISO does not provide certification itself; you would use an independent external certification body. We do not recommend or compare certifiers. Whichever body you consider, confirm its accreditation and its scope for this standard directly with it.

Do small businesses need ISO 42001?

ISO says the standard is intended for organisations of any size. Whether you need it depends on your AI use, your customers and your risk. A small business using a few low-risk tools may be better served by the government's Foundations guidance and a simple policy and register first.

Does ISO 42001 make us compliant with the Privacy Act?

No. The Privacy Act and the Australian Privacy Principles apply separately, and a management system does not guarantee compliance. It does give you the inventory, impact assessments and records you need to meet privacy obligations consistently, including the privacy policy requirement for certain automated decisions that starts on 10 December 2026.

How does ISO 42001 relate to the Voluntary AI Safety Standard?

The Voluntary AI Safety Standard names AS ISO/IEC 42001:2023 as the most important international standard it aligns with. The government's Guidance for AI Adoption, published on 21 October 2025, evolves that standard. ISO/IEC 42001 can be assessed independently; the Voluntary AI Safety Standard is voluntary and is not a certification scheme.

How long does implementation take?

It depends on scope, the number of AI systems, existing management systems, supplier dependence and ownership. We do not give a fixed number of weeks. A narrow initial scope that you run properly and then extend is usually the safest approach.

Sources

Official pages read on 31 October 2026:

Next step

List every AI tool and system your organisation uses today, including the informal ones, and mark which ones touch personal information or affect decisions about people. That list is the starting point for scope, risk and the privacy policy changes due on 10 December 2026. If you would like help turning it into a scoped ISO/IEC 42001 plan, book an AI governance consult and we will talk through your systems, your obligations and a sensible first scope.