What do Australian businesses need to do about AI and privacy by 10 December 2026?
'Should we have an AI policy?' comes up after almost every workshop I run, usually because a business down the road just published one. It's a fair question, and the document with a legal deadline attached is a different one, the privacy policy already sitting in your website footer. The date is 10 December 2026.
From that day, if your business is covered by the Privacy Act and uses software to make decisions about people, or to do something substantially and directly related to making them, and those decisions could significantly affect someone's rights or interests, your privacy policy has to say so. You list the kinds of personal information the software uses and the kinds of decisions it makes or shapes. Nothing becomes illegal on 10 December and you can keep using every tool you use now (it's a transparency rule, so most of the work is finding out what you've got).
What changes on 10 December 2026?
Three new clauses are added to Australian Privacy Principle 1, the principle that covers privacy policies. They came in through the Privacy and Other Legislation Amendment Act 2024, which received Royal Assent on 10 December 2024 and gave this part a two-year run-up. APP 1.7 says your policy needs the extra information when three things are true at once.
- You've arranged for a computer program to make a decision, or to do something substantially and directly related to making it.
- The decision could reasonably be expected to significantly affect an individual's rights or interests.
- Personal information about that person is used when the program runs.
A decision counts whether it helps or hurts the person, and refusing or failing to decide counts too. The Act gives three examples of decisions that can affect someone's rights or interests: granting or refusing a benefit under a law, a decision that affects their rights under a contract or agreement, and a decision that affects their access to a significant service or support.
Two details are easy to miss. The rule applies to decisions made after it starts, even if you set the software up years ago and the data was collected before then. And the OAIC published a fact sheet, a flowchart and updated APP 1 guidelines on 30 September 2026, shaped by 90 written submissions, so the regulator's view is already on paper.
Does this apply to my business?
It applies to APP entities, which is the Privacy Act's name for the organisations and Australian Government agencies it covers. Most businesses with an annual turnover of $3 million or less are exempt as small businesses, but the OAIC's small business page lists exceptions that pull smaller operators back in regardless of turnover. These include:
- health service providers, which covers allied health, complementary therapists, child care centres and private schools
- businesses that trade in personal information
- contractors providing services under a Commonwealth contract
- reporting entities under the anti-money laundering laws, for those activities
- a business related to a larger company that's covered
- any small business that has opted in.
Turnover means all income from all sources, and the test is whether you've gone over $3 million in any financial year since 2002, so one big year back in 2019 can be enough to put you in. (Read that one twice.)
What counts as solely or substantially automated?
Solely automated means the software makes the call and no person decides. Substantially automated is the bigger and blurrier group, where the program does something substantially and directly related to the decision, like scoring, ranking or recommending, and a person then decides with that output in front of them.
The law firm Allens reads the OAIC guidance as saying a tool counts when its output is a key factor with a direct connection to the decision, whether it advises or determines, and that human review doesn't take the tool out of scope (Allens). So 'a person signs off on it' won't get a recruitment screening tool off the hook.
Here's how I'd sort some everyday examples. Treat the last column as a starting point for a conversation with your adviser, because your facts will matter.
| Decision | Solely or substantially automated? | Affects rights or interests? | Disclose? |
|---|---|---|---|
| An online lender's system approves or declines a personal loan with no person involved | Solely | Yes, access to credit | Yes |
| Recruitment software ranks applicants and the manager only reads the top 20 | Substantially | Yes, employment | Yes |
| A tenant screening score the property manager relies on to choose an applicant | Substantially | Yes, access to housing | Yes |
| Gym software suspends a membership after two failed payments | Solely | Yes, rights under a contract | Likely yes |
| Claims software flags an insurance claim for a fraud check | Substantially | Likely, it can delay or deny the claim | Likely yes, get advice |
| A pricing tool shows different customers different prices based on their profile | Solely or substantially | Possibly | Get advice |
| A website chatbot answers questions about opening hours | No decision about the person | No | No |
| Staff use Claude to draft a newsletter or summarise meeting notes | No decision about a person | No | No |
The pricing row is there because Allens notes the guidance may capture targeted content that results in personalised pricing, and nobody has drawn that line cleanly yet.
What has to go in the privacy policy?
APP 1.8 asks for three things: the kinds of personal information used in those programs, the kinds of decisions made solely by them, and the kinds of decisions where the program does something substantially and directly related to the decision. It asks for kinds, so you don't have to publish your algorithm or name every vendor.
A plain-English paragraph is the goal. For a property manager, it might read something like this.
We use software to help assess rental applications. It uses the information in your application, including your income, rental history and identity documents, to produce a score. Our property managers use that score as a key factor when deciding which applicant to offer a lease to. We don't make any decision about your application solely by automated means.
That's an illustration of the shape, and your own wording has to match what your systems do.
What are people getting wrong about AI law in Australia?
I see the same three mix-ups in LinkedIn posts and sales decks most weeks.
There's no Australian AI Act on the way. The mandatory guardrails proposed in September 2024 never became law, and the National AI Plan released on 2 December 2025 says the government's approach builds on existing legal and regulatory frameworks, with each regulator handling AI harms in its own area and a new AI Safety Institute advising them. Privacy, consumer and workplace law all apply to AI already.
The ten guardrails are out of date. The Voluntary AI Safety Standard and its ten guardrails were followed on 21 October 2025 by the National AI Centre's Guidance for AI Adoption, which the government describes as updated and simplified guidance built on six essential practices. Those are: decide who is accountable, understand impacts and plan accordingly, measure and manage risks, share essential information, test and monitor, and maintain human control (National AI Centre). Both are voluntary.
The scary penalty numbers are wrong. If someone quotes an eight-figure fine at you about this deadline, they've grabbed the wrong number. The Act adds APP 1.7 to the list of privacy policy breaches the OAIC can deal with by infringement notice. On the OAIC's current figures, that's up to 60 penalty units per contravention for a company ($21,840) and 200 penalty units for a listed company ($72,800), and the OAIC can go to court for civil penalties if a notice isn't paid. That's real money for a small business, and a long way from the headlines.

What should I do in the next three weeks?
- Confirm whether you're an APP entity. If you're under $3 million, run the OAIC's small business checklist.
- List every tool that touches a decision about a customer, applicant, tenant, member or staff member. Include the automated features inside software you already pay for (CRMs, HR platforms, lending and booking systems), because that's where they tend to hide.
- For each one, write down the decision, what personal information goes in, and whether a person decides or the software does.
- Run each one through the three APP 1.7 tests. Anything that meets all three goes in the policy.
- Ask your vendors how their automated features work. The obligation sits with you as the business using the product, so get their answer in writing.
- Draft the new section of your privacy policy in plain English and have it reviewed. Anything in your 'get advice' column goes to a privacy lawyer.
- Publish it before 10 December, and set a calendar reminder to revisit it whenever you add a new tool.
This is general information and not legal advice. Your obligations depend on your business and your systems, so have a privacy lawyer check your wording before you publish.
Steps two and three take the longest, and they're also the most useful, because the same list shows you where AI is saving your team time and where it's making calls nobody has checked in a while. That mapping is where our Thread strategy sessions start. If you're also working out what staff can paste into ChatGPT in the first place, I've covered that in is it safe to put company data into ChatGPT?
And if your list turns out to be one chatbot and a newsletter tool, congratulations... you've got the shortest privacy policy update in Adelaide!
Questions people ask
Does using ChatGPT to write emails count as automated decision-making?
Usually not. The obligation covers decisions about a person that could significantly affect their rights or interests, such as access to credit, a job, housing or a significant service. Drafting emails, summarising notes or writing marketing copy doesn't decide anything about an individual, so it doesn't trigger the new privacy policy disclosure. The answer changes if you start using AI output to decide something about a specific person.
Our software vendor runs the automated decision. Is it their problem?
The obligation sits with the business that has arranged for the program to make or substantially shape the decision, which is you as the user of the product. Your vendor's own privacy policy doesn't cover your customers' decisions. Ask the vendor for a clear description of what their automated features do and what personal information they use, so you can describe it accurately in your policy.
Do we have to explain how the algorithm works?
No. APP 1.8 asks for the kinds of personal information used and the kinds of decisions made solely or substantially by the program. It does not ask you to publish the logic, the model or the vendor. The goal is a description an ordinary customer can follow, written in plain English and kept up to date when your systems change.
What happens if our privacy policy is wrong on 11 December?
Missing APP 1.7 information becomes one of the privacy policy breaches the OAIC can deal with by infringement notice, or by a compliance notice telling you to fix it. The OAIC weighs a range of factors before taking action. If you're mid-review on the day, finish the mapping, publish the best accurate version you can and keep a record of what you did and when.
Sources
- Federal Register of Legislation, Privacy and Other Legislation Amendment Act 2024 (as made)
- OAIC, New resources on transparency for use of AI and automated decision-making (30 September 2026)
- OAIC, Infringement notices
- OAIC, Small business
- Allens, Mandatory automated decision-making disclosure requirements are coming
- Department of Industry, Science and Resources, National AI Plan: Keep Australians safe
- Department of Industry, Science and Resources, Voluntary AI Safety Standard
- National AI Centre, Guidance for AI adoption: implementation guidance