From 10 December 2026, if your organisation is covered by the Privacy Act and software makes or shapes a decision that could significantly affect someone, your privacy policy has to say so. That includes AI. It can include a humble spreadsheet too.
Most privacy policies were written before anyone in the office had Copilot. So this one is going to sneak up on a lot of people.
The short version
- What: your privacy policy must explain when computer programs make, or feed into, decisions that could significantly affect people.
- When: 10 December 2026.
- Who: organisations covered by the Privacy Act. Most businesses over $3 million turnover, plus some smaller ones.
- Watch out for: having a human check the output doesn’t take a tool out of scope.
What actually changes on 10 December 2026?
The Privacy and Other Legislation Amendment Act 2024 (opens in a new tab) added three new rules to Australian Privacy Principle 1: APPs 1.7, 1.8 and 1.9 (opens in a new tab). They start on 10 December 2026 (opens in a new tab).
From that date, your privacy policy has to describe (opens in a new tab):
- the kinds of personal information used in the computer programs that make or support those decisions
- the kinds of decisions made by a program on its own
- the kinds of decisions where a program does something “substantially and directly related” (opens in a new tab) to making the decision
It’s a transparency rule. You don’t have to notify each person, and it doesn’t create a new right to challenge the decision (opens in a new tab). You do have to be upfront about what’s going on.
Does it apply to my organisation?
It applies to APP entities. In plain English that means Australian Government agencies and organisations with an annual turnover of more than $3 million (opens in a new tab).
Some smaller businesses are covered as well, whatever their turnover. The OAIC lists (opens in a new tab) private health service providers (including allied health, gyms and weight loss clinics), businesses that buy or sell personal information, child care centres, private schools and businesses that have opted in, among others.
If you’re under $3 million and none of those apply, the rule generally won’t catch you. It’s still a cracking habit to know where AI is making calls about your customers and staff.
What counts as a “computer program”?
More than you’d think. Allens’ reading of the OAIC’s final guidance (opens in a new tab) is that it covers rule-based tools, AI and machine learning, generative AI, chatbots and everyday software such as spreadsheets.
The rule kicks in when all three of these are true (APP 1.7 (opens in a new tab)):
- Your organisation has arranged for a program to make the decision, or to do something substantially and directly related to making it.
- The decision could reasonably be expected to significantly affect someone’s rights or interests. Allens points to health, housing, finance, employment and utilities (opens in a new tab).
- Personal information about that person is used in the operation of the program to make or support the decision.
If a human checks it, are we off the hook?
No. Allens’ reading of the guidance (opens in a new tab)is that human oversight doesn’t take a tool out of scope. If the AI’s output is a material input to the decision, assume it’s captured.
“Someone always checks it” sounds like a guardrail. On its own it isn’t one. Human-led means your people understand what the tool is doing and why, not just that someone clicks approve at the end.
Where AI hides in your decisions
You can’t put in your privacy policy what you don’t know is happening.
56% of Australian workers who use AI at work rely on tools their employer hasn’t approved (Salesforce and YouGov survey of 1,293 Australian adults, May 2026) (opens in a new tab). That’s the shadow AI problem, and it turns up in exactly the places this rule cares about.
A few illustrative examples (made up, but you’ll recognise them):
- A manager pastes 40 CVs into a free AI tool and asks for the top five.
- A customer service team uses a chatbot to sort hardship requests into “urgent” and “can wait”.
- A spreadsheet scores rental or membership applications, and nobody remembers who wrote the formula.
Your people are already using AI. The question is whether your privacy policy knows that.
Your 10 December check: six questions
Grab whoever looks after privacy, IT and HR. Give it an hour.
- What software touches decisions about people? List it all: AI tools, the vendor platforms, the spreadsheets.
- Which of those decisions could significantly affect someone? Jobs, money, housing, health, access to services.
- What personal information goes in? Name the kinds of information, not the systems.
- Is the program deciding, or feeding the decision? You’ll need to describe both kinds.
- What are your people using that you didn’t approve? Ask kindly. You’ll get a more honest answer.
- Does your privacy policy explain all of this in plain English? If not, draft the update and have your lawyer review it.
Bonus. Let your accounts find the software for you
Question 1 is the hard one, because nobody has a complete list. Your accounting software has most of it already: every subscription you pay for shows up as a transaction.
In Xero:go to Accounting, then Reports, then Account Transactions. Pick the accounts your software and subscriptions are coded to (often “Subscriptions”, “Software”, “IT” or “Computer expenses”), set the period to the last 12 months, then export to Excel. Do the same for expense claims, because that’s where tools paid on staff cards hide. MYOB and QuickBooks have similar transaction reports.
Before you paste the export into any AI tool, delete the columns you don’t need: staff names, card numbers, bank details and anything personal. Supplier names, descriptions and amounts are all you need.
Then use this prompt:
Then check the “unsure” rows with the person who uses each tool. One more thing: free AI tools won’t show up in your accounts at all, so question 5 still matters.
I'm not a lawyer
This is my plain-English take on a new rule under the Privacy Act 1988 (Cth), not legal advice. Every organisation is different, so check your own situation with your lawyer or privacy adviser, and read the OAIC’s guidance for yourself (links below).
What to do on Monday morning
- Book the hour. Run the six questions with privacy, IT and HR in the room.
- Ask your vendors one question: what does your tool decide, or feed into, about our customers or staff?
- Update the privacy policy and get it reviewed before 10 December.
- Then fix the bigger gap. A privacy policy update tells people what’s happening. An AI policy your people actually read decides what should happen. That’s the Guardrails step in GIST, and it always comes first.
Want your leadership team on the same page before the deadline?
A GIST Day takes your leadership team through Guardrails, Intent, Strategy and Practical Training in one day, with a written plan within 10 business days.
Booking for your team before Christmas? See the year-end team offers
Questions people ask
When does the automated decision-making rule start?
10 December 2026 (opens in a new tab). It’s in APPs 1.7 to 1.9, added to the Privacy Act by the Privacy and Other Legislation Amendment Act 2024.
Does it apply to small businesses?
Generally not if your annual turnover is $3 million or less (opens in a new tab), unless an exception applies. Private health service providers, businesses that trade in personal information, child care centres and private schools are among those covered regardless of size.
Is ChatGPT or Copilot a "computer program" under the rule?
It can be. If you use it to make, or substantially shape, a decision that could significantly affect someone, and personal information goes in, it’s likely captured.
Do we have to tell each person their decision was automated?
No, according to Allens’ reading of the OAIC guidance (opens in a new tab). It’s a transparency obligation in your privacy policy, not a requirement to notify people individually.
Does having a human review the output take it out of scope?
No, according to Allens’ reading of the OAIC guidance (opens in a new tab). If the program’s output is a material input to the decision, treat it as in scope and check with your adviser.
Where can I read the official guidance?
The Office of the Australian Information Commissioner (OAIC) has published a fact sheet, a flowchart and updated APP 1 guidelines. Links below.
Sources and further reading
- OAIC: New resources on transparency for use of AI and automated decision-making (opens in a new tab)
- OAIC: Rights and responsibilities under the Privacy Act (opens in a new tab)
- Allens: Mandatory automated decision-making disclosure requirements are coming (opens in a new tab)
- AICD and Human Technology Institute: A Director’s Guide to AI Governance, Version 2 (opens in a new tab)
- Salesforce: AI is widespread in Australian workplaces, but fragmentation is holding back impact (opens in a new tab)
Ada, my AI assistant, pulled the research together and shaped the structure. I checked every claim against the OAIC and legal guidance, and the opinions are mine. Human-led. AI-leveraged. How I work with AI
