On 10 December 2026 a new rule in the Privacy Act starts. If your business is covered by the Act and a computer program makes, or heavily shapes, decisions that matter to people, your privacy policy has to say so.
It has been reported as an AI rule. It is wider than that. The regulator reads "computer program" to take in a spreadsheet formula, a rule in your booking system and software you license from someone else, as well as a chatbot or an AI model. So the question for an owner is not "do we use AI?" but "does software decide things about people here?"
This post covers who the rule applies to, the three conditions that trigger it, what the policy has to state and what to do before the date. It is general information, not legal advice.
Key facts
- From 10 December 2026, Australian Privacy Principles 1.7 to 1.9 require a covered business to describe in its privacy policy how computer programs are used to make certain decisions about people.
- The rule was added to the Privacy Act 1988 by the Privacy and Other Legislation Amendment Act 2024.
- The Office of the Australian Information Commissioner (OAIC) published updated APP 1 guidelines, a fact sheet and a flowchart on the rule on 30 September 2026, about 10 weeks before it starts.
- Three conditions must all be met: a computer program makes or substantially and directly shapes a decision, the decision could significantly affect a person's rights or interests, and the program uses personal information about that person.
- The Privacy Act generally applies to organisations with annual turnover above $3 million, and to some smaller ones regardless of turnover, including health service providers and businesses that trade in personal information.
- The rule is about transparency only: Allens notes it gives people no new right to contest a decision and does not require you to notify each person directly.
Quick answer: if your turnover is $3 million or less and you are not in one of the excepted groups, the rule most likely does not apply to you, though it is still a sound habit to follow. If the Privacy Act does cover you, list every tool that uses personal information to make or shape a decision about a customer, applicant or staff member, and test each against the three conditions. For those that meet all three, add plain words to your privacy policy on what information is used and what kinds of decisions are made, by 10 December 2026. A person signing off at the end does not take a decision out of the rule.
Does the rule apply to my business?
The rule applies to what the Act calls APP entities. Start there, because it settles the question for many small businesses.
| Your business |
Covered by the Privacy Act? |
| Annual turnover above $3 million |
Yes |
| Turnover of $3 million or less, and you provide a health service |
Yes |
| Turnover of $3 million or less, and you trade in personal information |
Yes |
| Turnover of $3 million or less, and you are a reporting entity under anti-money laundering law |
Yes, for information handled for that purpose |
| Turnover of $3 million or less, and you work under a Commonwealth contract or have opted in |
Yes |
| Turnover of $3 million or less, and none of the above |
Generally exempt |
Checked on 16 November 2026 against the law firm summaries listed in the sources.
Two notes. Health service provider is read broadly: Lawpath's summary takes in allied health, physiotherapy and psychology, so a small clinic is covered whatever it turns over. And the small business exemption has been under review for removal, so exempt today does not mean exempt for good. The OAIC is the place to check your own position.
The three conditions
All three have to be met before the policy needs to change.
1. A computer program makes the decision, or does something substantially and directly related to it. The law firms that read the OAIC guidance agree the regulator takes a broad view of "computer program": simple rule-based processes, machine learning, everyday software including spreadsheets, and generative AI and chatbots. Squire Patton Boggs describes the test for "substantially and directly related" as the program's output being a key factor with a direct connection to the decision, and calls the threshold low.
2. The decision could reasonably be expected to significantly affect a person's rights or interests. Significant means more than trivial. The summaries point to decisions about employment, housing, credit, insurance, health services and the price of or access to essential goods and services. The effect can be good or bad for the person, and refusing or failing to decide counts as a decision. The guidance also looks at vulnerable people: an effect that is trivial for most can be significant for a child or a person with a disability.
3. Personal information about the person is used in running the program.
A human in the loop does not settle it. Piper Alderman puts it plainly: human review does not, on its own, remove the requirement. What matters is how much the person relies on the output, how often they override it and whether the output is advice or the answer.
Examples the law firms draw from the OAIC material and their own reading:
| Example |
In scope? |
| A spreadsheet that ranks patients for a health service |
Yes |
| Screening software that removes job candidates who lack a required degree |
Yes |
| An AI chat tool used to draft performance assessments that decide promotions |
Yes |
| Prices for essential goods that differ by postcode |
Yes |
| A fraud tool that freezes a customer's account |
Yes, on Lawpath's reading |
| A program whose output only informs a decision and is not a key factor in it |
No |
What the privacy policy has to say
For decisions that meet all three conditions, the policy must describe three things:
- The kinds of personal information used by those programs.
- The kinds of decisions made solely by a program.
- The kinds of decisions where a program does something substantially and directly related to the decision, even if a person approves it.
You do not have to publish your algorithm, your scoring logic or a list of every product you use. Similar decisions can be grouped, as long as the description still means something to an ordinary reader. Allens and Piper Alderman both note that sensitive inputs such as health or biometric information should be called out clearly, and that the commercial confidentiality carve-out is narrow: you can keep how a tool works to yourself, but not the fact that personal information goes into it.
Vendor software is still your disclosure
This is the part most likely to catch a small business. The obligation sits with the business that arranged for the program to be used, not the vendor that built it. If you bought, configured or switched on a tool that decides things about your customers, the disclosure is yours.
That matters for the AI tools businesses are putting in front of customers now:
- Customer service agents. A support agent such as Fin that answers questions about opening hours is not making a significant decision. Set the same agent up to approve or refuse refunds, claims or account changes from a customer's details and you may be inside the rule.
- Phone agents. A voice agent built on Vapi or Retell AI that takes messages is one thing. One that decides who gets an appointment, a call-back or a service is another.
- Assistants used on people. Asking ChatGPT to rank job applications or draft a staff assessment is the kind of use the law firms' examples describe.
The tool does not trigger the rule. The decision does. Ask each vendor, in writing, what their product decides and what personal information it uses. Allens recommends contracts say clearly who is making the decision.
What to do first
- Check whether the Privacy Act covers you, using the table above and the OAIC's own pages.
- List every tool that touches personal information, including spreadsheets and rules in your CRM, booking and billing systems. Note what each decides, about whom, and how often staff override it.
- Test each against the three conditions. Write down why a close call is in or out.
- Ask your vendors what their software decides and what data it uses.
- Add the three disclosures to your privacy policy in plain words, and have a lawyer review it.
- Publish before 10 December 2026, and recheck whenever you add a tool.
If you have a one-page AI policy for staff, this sits beside it. Our guide to writing an AI use policy for a small team covers the internal rules. The privacy policy is the public one.
Questions people ask
Does the automated decision-making rule only apply to AI?
No. It applies to any computer program that meets the three conditions. The law firms that reviewed the OAIC guidance say it covers rule-based tools, spreadsheets and everyday software as well as machine learning and generative AI. A formula that has ranked applicants for ten years can be covered.
Are small businesses exempt from the automated decision-making rule?
Many are, because the Privacy Act generally exempts businesses with annual turnover of $3 million or less. The exemption does not apply to health service providers, businesses that trade in personal information, and some other groups. It has also been under review for removal. Check your position with the OAIC.
Do I have to tell each customer that software made the decision?
Not under this rule. It is a privacy policy requirement. Allens notes it creates no right to contest a decision or to ask for an explanation, and no duty to notify people one by one. Other obligations, such as collection notices and consumer law, still apply.
Does a person checking the result take the decision out of the rule?
Not on its own. If the program's output is a key factor in the decision, it is covered even when a person signs off. How heavily staff rely on the output and how often they override it both count.
Is there a grace period after 10 December 2026?
None of the summaries we read describes one. Lawpath says there is no phase-in and no grace period, and that the regulator can issue infringement notices for privacy policy deficiencies and has run its own reviews of privacy policies.
Where to start
Sources checked on 16 November 2026: Allens, 2 October 2026 (the 30 September guidance, the three conditions, what the policy must contain, vendor software, the transparency-only point); Squire Patton Boggs, October 2026 (the breadth of "computer program", the key factor test, the patient ranking, candidate screening and postcode pricing examples); Piper Alderman, 20 October 2026 (the amending Act, who is an APP entity, human review, level of detail, the performance assessment example); Lawpath, 16 October 2026 (the small business exemption and its exceptions, the fraud tool example, no grace period, the steps). The OAIC's announcement, fact sheet and guidelines did not respond to our automated fetch, so the OAIC's position is given as these four firms report it; read the OAIC's own pages before you act.