One in four firms has named who is responsible for AI
The Australian Government’s Guidance for AI Adoption sets out six practices, and the first is deciding who is accountable. Most organisations have not done it. Here is what the guidance asks a firm of 10 to 200 people to put in writing.
Manish Kumar, founder · · 5 minute read
Australia has had an official answer to "what should a business do about AI" since 21 October 2025. It is called the Guidance for AI Adoption, it comes from the National AI Centre, and its short version is written for firms that are starting out. It asks for six practices. The first one is a name.
25 %
of organisations have clearly designated roles with responsibility for the responsible use of AI. It is among the least implemented of the 45 practices the index tracks.
Source: Fifth Quadrant, Australian Responsible AI Index 2025, 418 organisations with 20 or more employees
That survey covers organisations of 20 people and up, so it says nothing direct about the smallest firms. There is little reason to think they are ahead. Among firms of 20 to 99 people in the same index, 28 % scored 50 or more out of 100.
Agreeing with it is not the same as doing it
Nobody argues against responsible AI. The gap is between the nod and the document.
Most organisations agree with the principles. Far fewer have put the practices in place.
Share of organisations, Australian Responsible AI Index 2025
- Agreed with statements about ethical AI performance78 %
- Had implemented the relevant practices29 %
Source: National AI Centre, Guidance for AI Adoption: implementation guidance, summarising the 2025 Responsible AI Index
The National AI Centre's own survey of small and medium businesses points the same way. About half of the firms that use AI check its output before it reaches a customer, which is the most common safeguard. Telling customers that AI is in use, and giving them a way to raise a concern, lag well behind.
The six practices, in the government's words
These are the names and the order the guidance uses.
- 01Decide who is accountableAssign a senior leader as the overall owner of AI governance, and create an AI policy that sets out how the organisation will use AI responsibly.
- 02Understand impacts and plan accordinglyWork out who is affected by each use, and give those people a channel to contest an outcome.
- 03Measure and manage risksScreen each AI system and use case, so the ones that carry unacceptable risk or need more attention are flagged before they go live.
- 04Share essential informationCreate and maintain an AI register of every system and how it is used, and disclose your use of AI.
- 05Test and monitorAsk the supplier for proof, test before you deploy, monitor afterwards, and extend your data and cyber security practices to AI systems.
- 06Maintain human controlEnsure meaningful human oversight, and build in points where a person can override the system.
The guidance comes in two versions. Foundations is the short one, written for small and medium businesses, not-for-profits and anyone new to AI governance. Implementation guidance is the long one, for governance professionals and technical teams who build or customise AI systems or manage higher-risk uses. A firm of 40 people using Copilot and ChatGPT starts with Foundations and may never need the other.
What that means you write down
The guidance is blunt about paperwork: document every activity you carry out under these practices. For a firm of 10 to 200 people, that comes to a short list.
A named owner. One senior person, not a committee, with AI governance in their job.
An AI policy. The National AI Centre's own checklist says it should cover what AI can and cannot be used for, who approves higher-risk uses, what data staff can put into tools, when staff need to oversee AI use, how to report issues or misuse, and when the policy will be reviewed.
An AI register. A table of every AI system, what it is used for, who is accountable and how much governance it needs. The guidance says it should also cover AI embedded in other systems, such as human resources and customer engagement tools. That is where most firms come up short, because nobody chose those features. They arrived in an update.
A screening step. The Centre publishes seven screening questions. There are no scores. A yes to any of them means that use may need stricter controls.
A disclosure and a way to complain. Say where AI is used, and tell people how to contest an outcome that affects them.
Evidence of testing and of human review. A note of what was checked before launch, who looks at it now, and where a person can step in.
None of this is long. A policy that fits on two pages and a register with a dozen rows will do more than a 40-page framework nobody opens. The National AI Centre publishes free templates for the policy, the register and the screening questions, linked in the sources below.
It is guidance, not law. The law sits next to it.
The guidance is not binding. The National AI Plan, published on 2 December 2025, says the government's approach builds on existing laws and talks about voluntary measures and shared guidance. The guidance itself is the first update to the Voluntary AI Safety Standard of 2024, with ten guardrails condensed into six practices. If you worked to the old standard, the long version includes a crosswalk.
In July 2026 the Prime Minister announced mandatory Australian Standards for AI, with legislation aimed at early 2027. As announced, those are about large data centres and AI infrastructure. Nothing in the announcement puts new governance duties on an ordinary business that uses AI tools.
What is binding is the law you already work under, and one part of it is about to bite. From 10 December 2026, the Privacy Act requires a business it covers to say in its privacy policy where a computer program makes, or substantially and directly helps make, a decision that significantly affects a person. The Office of the Australian Information Commissioner published its final resources on 30 September 2026. Its guidelines say a decision can be in scope even where the program's output is subject to human review, and even where the program is supplied by someone else.
The register in practice four is how you find those decisions. A firm that follows the guidance has most of the Privacy Act work done as a side effect. A firm that does not has to do the inventory anyway, against a deadline.
Where to start this week
Pick the owner today. It costs nothing and it is the step three in four organisations have skipped.
Then start the register before the policy. List every tool with AI in it, including the features that arrived inside software you already pay for. A policy written before you know what is in use describes a business you do not have.
Then write the policy against the six-point checklist, run the seven screening questions on each row of the register, and book a date to review both.
The guidance aligns with the international standards a larger customer or insurer may ask about: the implementation version names AS ISO/IEC 42001:2023 and the NIST AI Risk Management Framework 1.0. That is alignment, not certification, but it means the short list above is the right first step whichever one you are asked about later.
Sources
- National AI Centre, Guidance for AI Adoption: foundations and implementation guidance
- National AI Centre templates: AI policy guide and template, AI systems register and AI screening questions
- Fifth Quadrant, Australian Responsible AI Index 2025, full report
- National AI Centre, AI adoption insights, December 2025 to February 2026
- Department of Industry, Science and Resources, Voluntary AI Safety Standard and National AI Plan: keep Australians safe
- Prime Minister of Australia, AI in Australia's interests, 15 July 2026
- OAIC, New resources on transparency for use of AI and automated decision-making, 30 September 2026 and APP Guidelines, Chapter 1
If you would rather hand this over: Guardrails Pack. Fixed scope · 10 working days · AU $2,900 ex GST. See the scope
Think you know the numbers? Guess eight published AI figures, then see each one with its source.