The short answer
A workable AI governance set-up for a business of 50 to 500 people comes down to 20 checks across five areas: ownership, your AI register, risk, disclosure and human oversight. You don't need a committee or a governance platform. You need one accountable executive, a short policy, a list of every AI system you use, and a few habits. Most businesses can tick off the first ten checks in a month.
On this page
Key takeaways
- Start with the register. You can't govern, disclose or test what you haven't listed.
- AI you buy counts as much as AI you build, including features switched on inside software you already use.
- Size the controls to the risk. An internal drafting tool and a loan scoring model need different checks.
- From 10 December 2026, some of this is a legal duty under the Privacy Act, not just good practice.
How should you use this checklist?
Go through each check and mark it done, partly done or not started. Be honest. A partly done check usually means it exists in someone's head but not in writing.
The checklist follows Australia's Guidance for AI Adoption, known as AI6, published by the National AI Centre in October 2025. AI6 is voluntary, but it is the yardstick customers, insurers and regulators now use. Work against it also counts towards ISO/IEC 42001, the international AI management standard.
Ownership and policy
- One named executive owns AI across the business. Written down, not assumed. IT runs the systems and legal advises, but this person answers for the outcomes.
- Every AI system has a named owner. Someone in the business who understands what it does and is accountable for how it is used.
- You have a short AI policy. Two pages covering which tools staff may use, what data can go into them, who approves a new tool, and what is not allowed.
- Staff know the policy exists. It has been shared, explained and is easy to find. A policy nobody has read is not a control.
Your AI register
- You have an AI register. A single list of every AI tool and feature in use. A spreadsheet is fine.
- It includes AI you didn't build. Vendor tools, AI features switched on inside software you already pay for, and anything staff signed up to themselves.
- Each entry records the basics. Owner, purpose, the data it uses, the decisions it touches, and a risk level.
- Someone checks it each quarter. New tools appear with every software update. The register is only useful if it is current.
Risk and testing
- New AI goes through a quick screen before it is switched on. Five questions, answered in writing: Does it use personal information? Does it make or shape decisions about people? Can a person check its output? What happens when it is wrong? Who owns it?
- Controls match the risk. Low-risk tools get light-touch rules. Anything that affects customers, staff or money gets more.
- AI is tested before go-live. Run it on real cases without letting it act, and compare its output with what your people would have done.
- Each live system has one measure you watch. Error rate, hours saved, complaints or turnaround. Someone looks at it regularly.
- You ask vendors for evidence. How they test their AI, what data it was trained on, and what happens to your data.
- AI is covered by your existing data and cyber security controls. Access, retention and incident response apply to AI tools too.
Disclosure and transparency
- You know which systems make or shape decisions about people. Run each one through the Privacy Act three-part test and write down the reasoning.
- Your privacy policy describes them. From 10 December 2026, businesses covered by the Privacy Act must explain the kinds of decisions software makes about people and the kinds of personal information it uses.
- People are told when they are dealing with AI. Wherever AI talks to customers or produces content that could be mistaken for a person's work.
Human oversight
- A person approves AI output where it matters. Anything that affects a customer, an employee or money has a human checkpoint.
- Someone can pause or switch it off. And critical work can continue if a system fails or a vendor changes something.
- There is a way to raise a problem. Staff and customers can question an outcome AI shaped, and someone investigates when something goes wrong.
How did you score?
- 16 to 20 done: You have a working set-up. Focus on keeping the register current and reviewing it every quarter.
- 8 to 15 done: The basics are partly there. Close the gaps in the register and disclosure first, especially before 10 December 2026.
- Under 8 done: Start with checks 1, 3 and 5: an owner, a policy and a register. Everything else depends on them.
Where should you start?
With the register. Four of the six AI6 practices depend on knowing what AI you have, and the Privacy Act disclosure can't be written without it. Most businesses find more AI than they expected, often switched on quietly inside tools they already pay for.
Governance also helps AI projects move. Pilots often stall when risk or legal ask who owns this and what happens when it's wrong. A current register and a short policy answer those questions in a day.
This article is general information, not legal advice. Work with your legal adviser on your privacy policy wording.
Frequently asked questions
What is an AI governance checklist?
A list of the controls a business needs to use AI safely and accountably: who owns it, what you have, how you manage risk, what you tell people, and how humans stay in control.
Do small and mid-sized businesses need AI governance?
Yes, but sized to fit. At 50 to 500 people, one owner, a two-page policy, an AI register and a quarterly review usually covers it. You don't need a committee or a dedicated platform.
Is AI governance a legal requirement in Australia?
AI6 is voluntary guidance. But from 10 December 2026, businesses covered by the Privacy Act must disclose decisions their software makes about people, and directors' duties already apply to decisions made with AI.
What is an AI register?
A list of every AI tool and feature your business uses, with its owner, purpose, data, the decisions it touches and its risk level. It is the foundation for both AI6 and Privacy Act disclosure.
How often should AI governance be reviewed?
Check the register and any incidents every quarter. Review the policy once a year, or sooner if you adopt a high-risk use of AI.
Do we need ISO 42001 certification?
Most mid-sized businesses don't. AI6 covers the essentials, and work done against it counts towards ISO 42001 if a major customer later asks for certification.
Not sure what AI you're already using?
Most businesses find more than they expected. Start your AI register.


