Contact Us
Contact Us
Australian Compliance

Australia's Automated Decision-Making Privacy Act Rules Start 10 December: What Your Software Must Disclose

From 10 December 2026, privacy policies must explain how software makes or shapes significant decisions about people. What is in scope, and how engineering can build the inventory.

JRC Software Team
8 min read
Australia's Automated Decision-Making Privacy Act Rules Start 10 December: What Your Software Must Disclose

A spreadsheet that ranks patients by need. A rules engine that declines a credit limit increase. A vendor tool that sorts job applicants before a recruiter opens the file. From 10 December 2026, any of these can put a new line in your privacy policy, because the automated decision-making Privacy Act amendments start applying on that date.

Most of these systems were never called "AI", so nobody thought to list them. CTOs and product managers keep asking us the same two things: which of our systems count, and what exactly do we have to write down? Scope is wider than most teams assume. It reaches rule-based logic, spreadsheets and vendor platforms as well as machine learning, and it can apply even when a person reviews the output.

Below: what the law requires, which software is likely caught, and the inventory your advisers need before they redraft the policy.

Note: This article is general information, not legal advice. Check the official guidance and get advice for your situation.

Where the 10 December 2026 date comes from

The obligation sits in Part 15 of Schedule 1 to the Privacy and Other Legislation Amendment Act 2024, headed "Automated decisions and privacy policies". The Act was assented to on 10 December 2024, and its commencement table starts Part 15 the day after a 24 month period beginning on that date, which the table lists as 10 December 2026.

The application provision is the part engineering teams tend to miss. The new rule applies to decisions made after commencement, whether the computer program was set up, and the personal information collected, before or after that date. Systems you built or bought years ago are in scope from day one.

The same Act adds APP 1.7 to the list of provisions in section 13K of the Privacy Act that can attract an infringement notice. A missing or vague disclosure becomes an enforceable gap, not a matter of drafting style.

On 30 September 2026 the Office of the Australian Information Commissioner (OAIC) announced new transparency resources, including a fact sheet, a flowchart, a supplementary fact sheet for government agencies and updated APP 1 Guidelines. Law firm commentary, including Allens' key takeaways and Squire Patton Boggs' analysis, reads the guidance as setting a low threshold.

What the automated decision-making Privacy Act rules require

APP 1.7 says your privacy policy must include extra information when all three of these are true:

  1. You've arranged for a computer program to make a decision, or to do something substantially and directly related to making it.
  2. The decision could reasonably be expected to significantly affect an individual's rights or interests.
  3. Personal information about that individual is used in the operation of the program.

APP 1.8 lists what to disclose: the kinds of personal information those programs use, the kinds of decisions made solely by a program, and the kinds of decisions where a program does something substantially and directly related to the decision. APP 1.9 adds that refusing or failing to decide counts as a decision, and that beneficial effects count as well as adverse ones.

Three readings of the OAIC guidance tend to surprise engineering teams:

  • "Computer program" is broad. Allens reports that it covers rule-based tools, AI and machine learning, everyday software including spreadsheets, and generative AI chatbots. Squire Patton Boggs describes an OAIC example where medical staff use a spreadsheet to rank people by medical need when scheduling appointments.
  • A human in the loop doesn't take you out. Squire Patton Boggs notes a decision may be in scope even where the program's output doesn't replace the whole process or is subject to human review. Advisory outputs that are a key factor in the decision can be caught.
  • Buying the software doesn't move the duty. Allens summarises the OAIC position as the disclosure obligation sitting with the user of a product, not the vendor.

Which systems are likely in scope

The OAIC examples go well past anything most teams would label "AI". The table maps software patterns we commonly see to the kinds of decisions the guidance describes. Treat it as a starting list, not a legal test.

System patternExample decisionWhy it may be caught
Recruitment and HR toolsSorting candidates, ranking employee performanceAffects access to employment, pay or promotion
Credit, lending and payment rulesApproving or declining applications, setting limitsAffects contractual rights and access to credit
Pricing enginesPersonalised or postcode-based pricingThe guidance treats small price differences on essentials as able to compound into a significant effect
Triage and eligibility scoringPrioritising service requests, health or support accessAffects access to a significant service or support
Fraud and risk modelsBlocking accounts, holding transactionsRefusing to act is itself a decision
Targeted job adsWho sees an employment opportunityAllens reports targeted employment advertising is expressly in scope

If you mapped automated decisions for Article 22 of the GDPR, don't assume that work covers you. Allens points out that Article 22 deals with decisions based solely on automated processing, whereas the Australian obligation is broader and also reaches programs that inform a human decision.

Building the inventory

Your privacy policy can only be as accurate as your inventory. We work from decisions back to systems, because starting from a list of applications misses spreadsheets and low-code flows every time.

  1. List significant decisions first. Ask each business unit which decisions affect customers, applicants, staff or members: approvals, refusals, pricing, prioritisation, suspensions, eligibility. Ignore the technology for now.
  2. Trace the software behind each one. Record which programs make the decision, recommend an outcome, score or rank inputs, or filter options. Include spreadsheets, rules engines, Power Automate or Zapier flows, and vendor SaaS.
  3. Record the personal information used. Note the fields each program reads. Allens reports the guidance expects more sensitive data, such as health or biometric information, to be called out more clearly than ordinary fields.
  4. Classify the program's role. Mark whether it decides alone or does something substantially and directly related to a human decision. Write down what human review really involves, including how often reviewers depart from the output, not what the process diagram claims.
  5. Assess significance. Note why the effect on the person is or isn't significant, with particular care for people experiencing vulnerability.
  6. Go to your vendors. Ask third-party providers how their tools use personal information in decisions, and check your contracts give you a right to that information. Squire Patton Boggs warns some suppliers may be reluctant to share the detail you need.
  7. Draft grouped, plain-language disclosures. Allens reads the guidance as permitting categorisation rather than listing every sub-decision. In our view a single line like "we may use automated tools" won't tell anyone the kinds of decisions or data involved, so it's unlikely to be enough.
  8. Give it an owner and a trigger. New features, new vendors and model changes should prompt an inventory update before release, not after.

Keep the inventory where engineering will maintain it. If you already run data catalogue or lineage tooling, our data governance starter framework overlaps heavily with steps 2 and 3.

One more point on confidentiality. Allens reports the OAIC's view that the commercially sensitive exclusion applies only where disclosure would genuinely harm commercial interests. Plan on describing the kinds of data and decisions even if you keep the model internals to yourself.

Where we fit in

We treat this as an engineering problem with a legal output. In the first week, our AI-driven software solutions team works through your codebase, data flows and vendor list to find each place a program makes, recommends or shapes a decision about a person. You get a structured inventory (decision, program, data used, role of the program, human review, vendor) that your privacy adviser can turn into policy wording. We don't write the policy or give legal advice, and we'll tell you not to wait for the inventory to be perfect before your adviser starts.

Where the inventory shows weak oversight, we help build the controls: logging model inputs and outputs, recording reviewer overrides, and adding the explanation and correction paths users expect. Our guide to UX patterns for AI features and trust covers the product side. For teams adding agents or models, we put a disclosure check into the release pipeline, as described in our post on starting small with AI agents.

Not sure which of your systems fall under the new rules? Talk to us about AI governance. We can help you build the inventory before 10 December 2026 and keep it current afterwards.

Frequently asked questions

Does APP 1.7 only apply to artificial intelligence?

No. APP 1.7 refers to a "computer program", and the OAIC reads that broadly. Commentary on the guidance says it covers rule-based processes, machine learning, generative AI, chatbots and everyday software such as spreadsheets. A formula that ranks people for access to a service can be in scope if it uses personal information and significantly affects them. Inventory all decision-support software, not just what your team calls AI.

If a person reviews every decision, are we out of scope?

Not necessarily. The obligation covers programs that do something substantially and directly related to a decision, not only programs that decide alone. Commentary on the OAIC guidance says a decision may be in scope even when a human reviews the output, particularly where that output is a key factor. Document what reviewers actually check and how often they depart from the program's result.

We use a vendor's platform. Is disclosure the vendor's job?

No. Using third-party software to support a decision doesn't shift the disclosure obligation to the vendor, even if you can't change the software's settings. You still need to describe the kinds of personal information used and the kinds of decisions affected. Ask vendors to explain how their tools use personal information, and check your contracts let you get that information when you need it.

Do we have to reveal how our models or algorithms work?

Generally not in technical detail. The legislation excludes commercially sensitive information, but the OAIC's view, as reported by Allens, is that the exclusion applies only where disclosure would genuinely harm your commercial interests. You still need to disclose the kinds of personal information used and the kinds of decisions affected. Get advice on where the line falls for your systems.

When do the automated decision-making rules start?

They commence on 10 December 2026, the day after the 24 month period that began when the Privacy and Other Legislation Amendment Act 2024 was assented to on 10 December 2024. They apply to decisions made after commencement, whenever the system was set up and whenever the personal information was collected, so existing systems need reviewing as well as new ones.

Australian Compliance

Talk to us about AI governance

Tell us where you're stuck and we'll come back with practical next steps.

Get in touch

Sources