Privacy Act Automated Decision-Making: Is Your Privacy Policy Honest About Your AI?

Privacy Act automated decision-making obligations commence on 10 December 2026, and they reach further than most leadership teams expect. The rule is simple to state: if software helps make significant decisions about people using their personal information, your privacy policy has to say so. The hard part is knowing where that is happening. Copilot, AI agents and years of embedded automation mean most organisations cannot answer that question today.

What Changes on 10 December 2026

The Privacy and Other Legislation Amendment Act 2024 added three new clauses to Australian Privacy Principle 1, which deals with open and transparent management of personal information. From 10 December 2026, APP entities must include information in their privacy policy where they have arranged for a computer program to make, or inform, decisions that could significantly affect individuals' rights or interests.

According to Gilbert + Tobin, the obligation applies to decisions made after commencement, even if the personal information was collected, or the system was already in use, before that date. Existing systems therefore need to be assessed, not just new ones.

On 30 September 2026 the Office of the Australian Information Commissioner (OAIC) released its final guidance: an updated APP 1 Guidelines chapter, a fact sheet, a supplementary fact sheet for government agencies and a decision flowchart. The OAIC says the material reflects 90 written submissions received during consultation. That left organisations roughly ten weeks to act.

It is worth being clear about what this is not. It is a transparency measure only. Unlike the EU GDPR, it does not give individuals a right to contest an automated decision, and it does not require you to notify individuals directly. What it does require is an accurate privacy policy.

The Three-Part Test for Privacy Act Automated Decision-Making

APP 1.7 sets out when disclosure is required. All three limbs must be met.

1. 01

   A computer program makes, or substantially and directly informs, a decision

   The entity has arranged for a computer program to make a decision, or to do a thing that is substantially and directly related to making it. Under APP 1.9, a decision includes refusing or failing to make one.
2. 02

   The decision could significantly affect an individual's rights or interests

   The effect can be adverse or beneficial. APP 1.9 gives examples: decisions about a benefit under legislation, decisions affecting rights under a contract or arrangement, and decisions affecting access to a significant service or support.
3. 03

   Personal information is used in the program's operation

   Personal information about the individual is used by the program to make the decision, or to do the thing that is substantially and directly related to it.

The OAIC's APP 1 Guidelines give practical colour to limb two. Examples of significant effects include access to healthcare, emergency or aged care services, housing, employment opportunities, essential utilities and telecommunications. The guidelines also note that effects may be more significant for vulnerable people, naming children, people from low socio-economic backgrounds, Indigenous Australians and people with disability. Targeted advertising may have a significant effect where it leads to personalised pricing or limits access to employment opportunities.

What Your Privacy Policy Must Now Disclose

Where the test is met, APP 1.8 requires the privacy policy to cover three things:

• ✓
  The kinds of personal information used in the operation of those computer programs.
• ✓
  The kinds of decisions made solely by the operation of those programs.
• ✓
  The kinds of decisions where a program does something substantially and directly related to making the decision.

The guidance is not prescriptive about format. The OAIC asks for clear, plain language and warns against excessive detail that obscures the purpose of transparency. Decisions and information can be grouped into categories, provided the result is meaningful to a reasonable person. Where more sensitive information is used, such as health information or biometric templates, the policy should say so clearly. Legal commentators note that generic, high-level statements are unlikely to be enough.

Why Copilot and AI Agents Bring This Close to Home

The OAIC's guidelines say "computer program" should be read broadly. It covers rule-based processes, AI and machine learning, commonly used software, apps and word-processing tools, and generative AI tools. The Explanatory Memorandum to the 2024 Act also refers to simple tools such as Excel. Microsoft 365 Copilot, Copilot Studio agents and third-party AI features are therefore candidates for assessment. Whether a given use is in scope depends on how it is used, not on the brand of the tool. Using software only to record a decision a human has already made is treated differently from using it to help make the decision.

Consider a few patterns we see as Copilot adoption matures. A hiring manager asks Copilot to compare candidate CVs against selection criteria and shortlist the strongest. A Copilot Studio agent triages customer hardship or complaint requests and routes some for priority handling. A finance team uses a spreadsheet model to decide which accounts go to collections. Each involves personal information, and each may materially shape an outcome for a person. Whether any of them meets the test is a question of fact, which is exactly why they need to be found and assessed.

The difficulty is visibility. Depending on how it is governed, Copilot Studio can let business users build agents outside formal IT projects, and AI features are increasingly built into HR, CRM and finance platforms. If you do not have an inventory of agents and AI-enabled workflows, you cannot write an accurate privacy policy.

The vendor question

The OAIC expects the obligation to remain with the entity that uses personal information to make the decision, even where another party supplies or operates the program. If your recruitment platform, CRM or insurance system has automated scoring built in and you use it to make decisions about people, assume the disclosure is yours. The guidelines say entities should keep oversight of third-party tools during and after procurement, and that contracts should state which entity holds ultimate decision-making responsibility.

Where Automated Decisions Hide in a Typical Business

ADM is rarely labelled as such. It tends to sit inside routine processes across HR, finance, marketing, customer service and operations. These are the areas we recommend every organisation reviews first.

A Four-Step Plan Before 10 December

With around four weeks to go, the aim is a defensible position on commencement day, not perfection. These four steps get you there.

1. 01

   Build the inventory

   List every computer program used in decisions about individuals: core systems, spreadsheets and models, vendor SaaS features, Copilot use cases and Copilot Studio agents. On the Microsoft side, the Power Platform admin center and Microsoft Purview audit logs can help show which agents exist and how Copilot is being used.
2. 02

   Apply the three-limb test to each use

   For each item, record the decision, the personal information used, the likely effect on the individual and how much people rely on the output. Use the OAIC flowchart and document your reasoning, including for items you conclude are out of scope.
3. 03

   Close the vendor gaps

   Where in-scope ADM sits inside third-party software, ask vendors what the feature does and what data it uses. Check that contracts make clear who is making the decision.
4. 04

   Update the privacy policy and make it repeatable

   Draft the APP 1.8 disclosures in plain language, have them reviewed by your legal adviser and publish before 10 December. Then add an ADM check to your AI governance process so new agents and tools are assessed before they go live, not after.

Penalties, Enforcement and What Comes Next

Privacy policies are already on the regulator's radar. In December 2025 the OAIC announced a compliance sweep focused on privacy policies, and the 2024 reforms gave it new enforcement tools, including infringement notices and compliance notices.

Two other developments sit alongside this deadline. The OAIC is required to register the Children's Online Privacy Code by 10 December 2026. It is aimed at online services likely to be accessed by children, and its commencement date and any transition period should be checked on the OAIC website. Separately, the Attorney-General's Department released an exposure draft of the second tranche of Privacy Act reforms on 31 August 2026, with submissions closing on 18 September. The final content and timing of those reforms are yet to be settled.

The practical lesson is that an ADM inventory is not a one-off compliance exercise. It is the foundation for whatever comes next in privacy and AI regulation, and it is the same asset you need to govern Copilot properly.

Who Needs to Act on Automated Decision-Making Now

The obligation applies to APP entities: Australian Government agencies and private sector organisations covered by the Privacy Act. In general that means businesses with annual turnover above $3 million, plus some smaller businesses, such as health service providers, that are covered regardless of turnover. If you are unsure whether you are covered, get advice.

The organisations with the most exposure are those making high-volume decisions about customers or staff: financial services and lending, insurance, healthcare, utilities, education, recruitment and government. Professional services firms rolling out Copilot across HR and client operations should also assess their position. It applies equally to organisations that have already deployed Copilot and agents and now need to look back at what is running.